Combine GA4 and ChatGPT To Deliver More Insightful Analytics [Step-by-Step Instructions]

Optimal day: Friday

Optimal topics: Content marketing, SEO

Start date: Upcoming Friday

Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Report: Acquisition > Traffic acquisition: Search default channel group

Dimension: Session campaign

Secondary dimension: Date

Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

“Based on this data, what type of campaigns should be sent more often?”

“What types of campaigns should be abandoned?”

“Suggest changes that would improve the efficiency of this email program.”

“Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).

Reevaluate SEO campaigns.

Enhance personalization.

Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.

Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.

SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.

Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

You’ve used the GA4 version of Google Analytics for at least a year, if not two to three. You’ve also probably tested the potential of ChatGPT and similar generative AI tools.

But have you combined the power of both?

That’s where Andy Crestodina comes in. The co-founder and CMO of Orbit Media Studios discussed the opportunities in his Marketing Analytics & Data Science (MADS) conference presentation, How to Audit Your Content Using GA4: A Data-Driven Approach to SEO and Lead Generation.

In this article, he walks you through step by step how to create GA4 reports and prompt ChatGPT to identify rising and falling search traffic, improve content rankings, and deliver more effective newsletters.

Which content has rising and falling search traffic?

To see traffic from organic search:

• Go to the Engagement section in the left column.
• In the dropdown, click on the report “Pages and screens: Page title and screen class.”
• Next, click “Add filter” to see organic traffic.

To build the filter:

• Go to the Include dropdown in the right column.
• Select Session medium.
• Under Dimension values, select organic.

At the top of the page, click on “Add comparison.” Select the period, such as the last seven days or the last 28 days. Scroll down to turn on the Compare toggle.

Andy says GA4’s date comparison feature is better than the old Universal Analytics because its comparison matches the days of the week between the two periods.

The resulting report identifies rising and falling search traffic as illustrated in this report:

In the report above, the URL — /blog/how-to-write-testimonials-examples — saw a 43.43% drop in views when comparing Jan. 7 through Feb. 3, 2023, to Dec. 10, 2022, through Jan. 6, 2023. But the URL — /blog/website-navigation — saw a 42.86% rise in search traffic during that same period.

Once you’ve identified content with falling search traffic, consider improving it with these ideas:

What phrases does this article rank for?

Knowing the phrases an article ranks for helps you understand why people find the page. It also helps determine how to optimize for additional phrases and drive more organic search traffic to it.

To do this, Andy enlists the help of ChatGPT. However, the first step is to generate a report from Google Search Console:

• Search type: Web
• Date: [a timeframe] (In this example, Andy selected the last 3 months.)
• Query: -orbit (i.e., to exclude branded queries for Andy’s agency Orbit Media)
• Page: [the page URL]

Click “Export” in the upper right to generate a CSV file containing the data.

Feed the CSV file into ChatGPT. Andy uses ChatGPT Plus ($20 a month) because it supports data analysis and file uploads. He starts with this prompt:

“I’m giving you Google Search Console data showing the search performance of a URL. Can you analyze?”

ChatGPT replies with the five columns of data and suggested areas it could analyze (e.g., queries generating the most clicks).

Since any ranking page also ranks for a bunch of other things, he consolidates the list with this prompt:

“You are an expert SEO who is highly proficient at keyword analysis. Merge the rows with similar queries into single rows with a single, representative marketing keyphrase. As you do so, for each row, combine the data for impressions and clicks and averaging the data for CTR and Avg Position. Provide link to download.”

Any time you ask AI to manipulate data, ask it to provide a link to download the data so you can confirm that it processed the data properly. In this instance, Andy finds the data from ChatGPT looks good.

The chart includes 19 rows of queries related to what should be on a website homepage, accompanied by their total clicks, impressions, CTR, and position.

With the data on what phrases the page ranks for, Andy pastes the content into ChatGPT and makes this prompt:

“Here’s the page for the data you just analyzed. Evaluate this content’s keyword usage and frequency based on the keywords in the dataset.”

ChatGPT replies with a set of steps it will perform, including text preprocessing and keyword extraction:

• Text preprocessing: We’ll clean the webpage to prepare it for analysis.
• Keyphrase extraction: We’ll identify the keyphrases that appear in the content.
• Frequency count: We’ll count how often each keyphrase appears in the content.
• Evaluation: Finally, we’ll assess the usage of keyphrases based on their frequency in the content and their importance in the aggregated Google Search Console data.

Andy refines the prompt to have ChatGPT get more sophisticated in its recommendations (bolded emphasis by Andy):

“Suggest edits to this page that would rank higher by better indicating its relevance for the keyphrases in the dataset. Focus on recommendations for including the phrases that are not used on the page but appear in the dataset. Make suggestions that improve the flow and clarity of the content, as well as keyphrase relevance. Highlight the recommended changes.”

What content edits can improve rankings?

I find the results quite amazing. Some of the recommended changes generated by ChatGPT:

Introduction

Current: Social media streams are crowded, fast-flowing, and very competitive. As social networks pull back on organic reach, showing up is hard, and getting attention is even harder.

Suggested edit: Navigating the crowded and competitive social media video landscape can be daunting. With platforms like Facebook and Instagram curtailing organic reach, the challenge of making your DIY social media videos visible is greater than ever.

Second section: Winning the algorithm

Current: Social media videos are so engaging that the networks are pushing them way to the top of social streams. Facebook, Instagram, and LinkedIn all want us to do more in this all-powerful format. It wins in the algorithms and it wins for attention.

Suggested edit: The power of creating engaging social media videos is so compelling that platforms like Facebook, Instagram, and LinkedIn are giving them preferential treatment. This is why learning how to create captivating social media video content is crucial—it wins both in algorithms and in capturing audience attention.

Third section: Tools and equipment

Current: Your laptop has a camera and mic, but you can do better than that. A little more effort will get you a high-quality video and much better results.

Suggested edit: While your laptop’s built-in camera and mic may suffice for starters, investing in specialized video equipment for social media marketing can significantly boost the quality and engagement of your DIY social media videos.

Which articles convert email subscribers the best?

Over the past 18 years, Andy has written over 575 articles on topics ranging from influencer marketing to web analytics to email marketing. Some articles perform amazingly, while others receive no traffic. Similarly, some articles get people to subscribe to Orbit’s email list, while others generate zero subscribers.

“When you know that difference, you know exactly what to do on social media. You know what to promote and what’s working well,” Andy says.

In the past, he used Excel spreadsheets to analyze the ­­ for every article on his site. These days, he uses ChatGPT.

First, he collects the data from GA4. Under the Engagement section, he selects “Page and screens: Page path + query string” report. To customize it, he clicks on the pencil icon in the upper right and adds these metrics:

• Sessions
• Users
• Conversions

TIP: If you don’t see your metric right away, type its name and select it when it appears.

Since Andy only wanted blogs in his data set, he clicks “Add filter” in the upper left and builds the filter with these attributes:

• Dimension: Page path and screen class
• Match type: contains
• Value: blog

From here, click the export icon in the upper right (i.e., it’s two icons to the left of the pencil icon) and select “Download CSV.”

Next, Andy creates a second CSV file by changing “Value: blog” to “Value: /blog-newsletter-thank-you-page” to collect his email signup thank you page data:

He clicks the “+” button left of the report columns to select a secondary dimension of “Page Referrer.” The resulting report shows the pages users read right before they subscribed to Orbit’s email list. He exports this data as a CSV.

Andy cleans up this file by deleting the nine rows of metadata at the top, so ChatGPT won’t get confused in its analysis.

The two files are fed into ChatGPT to confirm it could analyze them. Next, Andy asks ChatGPT to do the equivalent of an Excel VLOOKUP so users can find things in a table or a range by row. He uses this prompt:

“Map the data from the two files together into one file, as you would do with vlookup, so each row is a URL and the sessions and conversions are shown for each.”

He asks ChatGPT to provide a file to download. ChatGPT generates 17 rows of URLs accompanied by sessions (traffic), users (traffic), conversions, sessions, users (converted), and conversions (newsletter):

Then, Andy prompts ChatGPT to draw a bar chart showing the conversion rate for each blog post.

ChatGPT responds with how it calculates the conversion rate for each blog post and visualizes a bar chart showing the top 20 blog posts by conversion rate.

NOTE: Conversion rate data is only available because ChatGPT divided the conversion data (numerator) by session traffic (denominator).

Next steps

To amplify the articles that drive the most email conversions, Andy suggests:

• Promoting the article(s) in your email signature.
• Featuring them on your homepage.
• Putting them into a social media marketing rotation.
• Linking to them in presentations.
• Linking to them from high-traffic pages.

Which email campaigns perform the best?

You can also use the GA4-ChatGPT combination to understand which email campaigns perform the best. Here’s how Andy does it:

• Export GA4 data showing traffic to pages based on UTM codes related to email campaigns.
• Export data from Orbit’s email marketing platform (ActiveCampaign).
• Ask ChatGPT to merge the two datasets, analyze, and provide recommendations

To collect the GA4 data, Andy uses the traffic acquisition report, which can be found under Acquisition in the left navigation bar. He clicks on the pencil in the upper right to customize it and selects the following metrics before clicking the “apply” button:

• Sessions
• Engagement rate
• Average engagement time per session
• Conversions
• Session conversion rate

Next, he clicks “Add filter,” selects the “Session medium” dimension, then the value of “email.”

Andy changes the primary dimension to “Session campaign.” The resulting report has one row for each email campaign (i.e., traffic statistics based on the utm_campaign parameter). Andy exports this data as a CSV file.

Next, Andy exports a data file from ActiveCampaign, which includes:

• Campaign name
• Number of sends
• Number of opens
• Number of clicks
• Number of unsubscribes
• Number of bounces
• Open rate
• Click rate
• Unsubscribe date
• Bounce rate

Andy uploads each file separately to ChatGPT and confirms it can process each file. He then prompts ChatGPT to merge the datasets:

“These two datasets show the performance of the same campaigns from both the email service provider (active campaign) and the website behavior (google analytics). I want to see the performance of each campaign across both datasets. Merge the Google Analytics data with the ActiveCampaign data, combining ‘session campaign’ with ‘campaign name’ into one column, inferring which campaigns were the same based on names and dates. Provide link to download.”

Once ChatGPT merges the datasets, Andy confirms the data looks good and puts ChatGPT to work.

The basic prompt is:

“You are a skilled digital strategist and email marketing expert. What insights can you gather from the combined dataset?”

Moonshot prompts include:

• “What can you tell me about newsletter campaigns?”
• “What insights can you gather from this data?”
• “Make a chart using an unusual visualization method showing unexpected insights from the data in this dataset.”
• “How would Leonardo da Vinci visualize this data? Show your thinking.”

Next, Andy asks ChatGPT to categorize the newsletter campaigns and provide insights on metrics for each category. It returns these strategic insights:

• Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.
• Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.
• SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.
• Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Report: Acquisition > Traffic acquisition: Search default channel group

Dimension: Session campaign

Secondary dimension: Date

Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

You’ve used the GA4 version of Google Analytics for at least a year, if not two to three. You’ve also probably tested the potential of ChatGPT and similar generative AI tools.

But have you combined the power of both?

That’s where Andy Crestodina comes in. The co-founder and CMO of Orbit Media Studios discussed the opportunities in his Marketing Analytics & Data Science (MADS) conference presentation, How to Audit Your Content Using GA4: A Data-Driven Approach to SEO and Lead Generation.

In this article, he walks you through step by step how to create GA4 reports and prompt ChatGPT to identify rising and falling search traffic, improve content rankings, and deliver more effective newsletters.

Which content has rising and falling search traffic?

To see traffic from organic search:

• Go to the Engagement section in the left column.
• In the dropdown, click on the report “Pages and screens: Page title and screen class.”
• Next, click “Add filter” to see organic traffic.

To build the filter:

• Go to the Include dropdown in the right column.
• Select Session medium.
• Under Dimension values, select organic.

At the top of the page, click on “Add comparison.” Select the period, such as the last seven days or the last 28 days. Scroll down to turn on the Compare toggle.

Andy says GA4’s date comparison feature is better than the old Universal Analytics because its comparison matches the days of the week between the two periods.

The resulting report identifies rising and falling search traffic as illustrated in this report:

In the report above, the URL — /blog/how-to-write-testimonials-examples — saw a 43.43% drop in views when comparing Jan. 7 through Feb. 3, 2023, to Dec. 10, 2022, through Jan. 6, 2023. But the URL — /blog/website-navigation — saw a 42.86% rise in search traffic during that same period.

Once you’ve identified content with falling search traffic, consider improving it with these ideas:

What phrases does this article rank for?

Knowing the phrases an article ranks for helps you understand why people find the page. It also helps determine how to optimize for additional phrases and drive more organic search traffic to it.

To do this, Andy enlists the help of ChatGPT. However, the first step is to generate a report from Google Search Console:

• Search type: Web
• Date: [a timeframe] (In this example, Andy selected the last 3 months.)
• Query: -orbit (i.e., to exclude branded queries for Andy’s agency Orbit Media)
• Page: [the page URL]

Click “Export” in the upper right to generate a CSV file containing the data.

Feed the CSV file into ChatGPT. Andy uses ChatGPT Plus ($20 a month) because it supports data analysis and file uploads. He starts with this prompt:

“I’m giving you Google Search Console data showing the search performance of a URL. Can you analyze?”

ChatGPT replies with the five columns of data and suggested areas it could analyze (e.g., queries generating the most clicks).

Since any ranking page also ranks for a bunch of other things, he consolidates the list with this prompt:

“You are an expert SEO who is highly proficient at keyword analysis. Merge the rows with similar queries into single rows with a single, representative marketing keyphrase. As you do so, for each row, combine the data for impressions and clicks and averaging the data for CTR and Avg Position. Provide link to download.”

Any time you ask AI to manipulate data, ask it to provide a link to download the data so you can confirm that it processed the data properly. In this instance, Andy finds the data from ChatGPT looks good.

The chart includes 19 rows of queries related to what should be on a website homepage, accompanied by their total clicks, impressions, CTR, and position.

With the data on what phrases the page ranks for, Andy pastes the content into ChatGPT and makes this prompt:

“Here’s the page for the data you just analyzed. Evaluate this content’s keyword usage and frequency based on the keywords in the dataset.”

ChatGPT replies with a set of steps it will perform, including text preprocessing and keyword extraction:

• Text preprocessing: We’ll clean the webpage to prepare it for analysis.
• Keyphrase extraction: We’ll identify the keyphrases that appear in the content.
• Frequency count: We’ll count how often each keyphrase appears in the content.
• Evaluation: Finally, we’ll assess the usage of keyphrases based on their frequency in the content and their importance in the aggregated Google Search Console data.

Andy refines the prompt to have ChatGPT get more sophisticated in its recommendations (bolded emphasis by Andy):

“Suggest edits to this page that would rank higher by better indicating its relevance for the keyphrases in the dataset. Focus on recommendations for including the phrases that are not used on the page but appear in the dataset. Make suggestions that improve the flow and clarity of the content, as well as keyphrase relevance. Highlight the recommended changes.”

What content edits can improve rankings?

I find the results quite amazing. Some of the recommended changes generated by ChatGPT:

Introduction

Current: Social media streams are crowded, fast-flowing, and very competitive. As social networks pull back on organic reach, showing up is hard, and getting attention is even harder.

Suggested edit: Navigating the crowded and competitive social media video landscape can be daunting. With platforms like Facebook and Instagram curtailing organic reach, the challenge of making your DIY social media videos visible is greater than ever.

Second section: Winning the algorithm

Current: Social media videos are so engaging that the networks are pushing them way to the top of social streams. Facebook, Instagram, and LinkedIn all want us to do more in this all-powerful format. It wins in the algorithms and it wins for attention.

Suggested edit: The power of creating engaging social media videos is so compelling that platforms like Facebook, Instagram, and LinkedIn are giving them preferential treatment. This is why learning how to create captivating social media video content is crucial—it wins both in algorithms and in capturing audience attention.

Third section: Tools and equipment

Current: Your laptop has a camera and mic, but you can do better than that. A little more effort will get you a high-quality video and much better results.

Suggested edit: While your laptop’s built-in camera and mic may suffice for starters, investing in specialized video equipment for social media marketing can significantly boost the quality and engagement of your DIY social media videos.

Which articles convert email subscribers the best?

Over the past 18 years, Andy has written over 575 articles on topics ranging from influencer marketing to web analytics to email marketing. Some articles perform amazingly, while others receive no traffic. Similarly, some articles get people to subscribe to Orbit’s email list, while others generate zero subscribers.

“When you know that difference, you know exactly what to do on social media. You know what to promote and what’s working well,” Andy says.

In the past, he used Excel spreadsheets to analyze the ­­ for every article on his site. These days, he uses ChatGPT.

First, he collects the data from GA4. Under the Engagement section, he selects “Page and screens: Page path + query string” report. To customize it, he clicks on the pencil icon in the upper right and adds these metrics:

• Sessions
• Users
• Conversions

TIP: If you don’t see your metric right away, type its name and select it when it appears.

Since Andy only wanted blogs in his data set, he clicks “Add filter” in the upper left and builds the filter with these attributes:

• Dimension: Page path and screen class
• Match type: contains
• Value: blog

From here, click the export icon in the upper right (i.e., it’s two icons to the left of the pencil icon) and select “Download CSV.”

Next, Andy creates a second CSV file by changing “Value: blog” to “Value: /blog-newsletter-thank-you-page” to collect his email signup thank you page data:

He clicks the “+” button left of the report columns to select a secondary dimension of “Page Referrer.” The resulting report shows the pages users read right before they subscribed to Orbit’s email list. He exports this data as a CSV.

Andy cleans up this file by deleting the nine rows of metadata at the top, so ChatGPT won’t get confused in its analysis.

The two files are fed into ChatGPT to confirm it could analyze them. Next, Andy asks ChatGPT to do the equivalent of an Excel VLOOKUP so users can find things in a table or a range by row. He uses this prompt:

“Map the data from the two files together into one file, as you would do with vlookup, so each row is a URL and the sessions and conversions are shown for each.”

He asks ChatGPT to provide a file to download. ChatGPT generates 17 rows of URLs accompanied by sessions (traffic), users (traffic), conversions, sessions, users (converted), and conversions (newsletter):

Then, Andy prompts ChatGPT to draw a bar chart showing the conversion rate for each blog post.

ChatGPT responds with how it calculates the conversion rate for each blog post and visualizes a bar chart showing the top 20 blog posts by conversion rate.

NOTE: Conversion rate data is only available because ChatGPT divided the conversion data (numerator) by session traffic (denominator).

Next steps

To amplify the articles that drive the most email conversions, Andy suggests:

• Promoting the article(s) in your email signature.
• Featuring them on your homepage.
• Putting them into a social media marketing rotation.
• Linking to them in presentations.
• Linking to them from high-traffic pages.

Which email campaigns perform the best?

You can also use the GA4-ChatGPT combination to understand which email campaigns perform the best. Here’s how Andy does it:

• Export GA4 data showing traffic to pages based on UTM codes related to email campaigns.
• Export data from Orbit’s email marketing platform (ActiveCampaign).
• Ask ChatGPT to merge the two datasets, analyze, and provide recommendations

To collect the GA4 data, Andy uses the traffic acquisition report, which can be found under Acquisition in the left navigation bar. He clicks on the pencil in the upper right to customize it and selects the following metrics before clicking the “apply” button:

• Sessions
• Engagement rate
• Average engagement time per session
• Conversions
• Session conversion rate

Next, he clicks “Add filter,” selects the “Session medium” dimension, then the value of “email.”

Andy changes the primary dimension to “Session campaign.” The resulting report has one row for each email campaign (i.e., traffic statistics based on the utm_campaign parameter). Andy exports this data as a CSV file.

Next, Andy exports a data file from ActiveCampaign, which includes:

• Campaign name
• Number of sends
• Number of opens
• Number of clicks
• Number of unsubscribes
• Number of bounces
• Open rate
• Click rate
• Unsubscribe date
• Bounce rate

Andy uploads each file separately to ChatGPT and confirms it can process each file. He then prompts ChatGPT to merge the datasets:

“These two datasets show the performance of the same campaigns from both the email service provider (active campaign) and the website behavior (google analytics). I want to see the performance of each campaign across both datasets. Merge the Google Analytics data with the ActiveCampaign data, combining ‘session campaign’ with ‘campaign name’ into one column, inferring which campaigns were the same based on names and dates. Provide link to download.”

Once ChatGPT merges the datasets, Andy confirms the data looks good and puts ChatGPT to work.

The basic prompt is:

“You are a skilled digital strategist and email marketing expert. What insights can you gather from the combined dataset?”

Moonshot prompts include:

• “What can you tell me about newsletter campaigns?”
• “What insights can you gather from this data?”
• “Make a chart using an unusual visualization method showing unexpected insights from the data in this dataset.”
• “How would Leonardo da Vinci visualize this data? Show your thinking.”

Next, Andy asks ChatGPT to categorize the newsletter campaigns and provide insights on metrics for each category. It returns these strategic insights:

• Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.
• Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.
• SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.
• Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

“Based on this data, what type of campaigns should be sent more often?”

“What types of campaigns should be abandoned?”

“Suggest changes that would improve the efficiency of this email program.”

“Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

You’ve used the GA4 version of Google Analytics for at least a year, if not two to three. You’ve also probably tested the potential of ChatGPT and similar generative AI tools.

But have you combined the power of both?

That’s where Andy Crestodina comes in. The co-founder and CMO of Orbit Media Studios discussed the opportunities in his Marketing Analytics & Data Science (MADS) conference presentation, How to Audit Your Content Using GA4: A Data-Driven Approach to SEO and Lead Generation.

In this article, he walks you through step by step how to create GA4 reports and prompt ChatGPT to identify rising and falling search traffic, improve content rankings, and deliver more effective newsletters.

Which content has rising and falling search traffic?

To see traffic from organic search:

• Go to the Engagement section in the left column.
• In the dropdown, click on the report “Pages and screens: Page title and screen class.”
• Next, click “Add filter” to see organic traffic.

To build the filter:

• Go to the Include dropdown in the right column.
• Select Session medium.
• Under Dimension values, select organic.

At the top of the page, click on “Add comparison.” Select the period, such as the last seven days or the last 28 days. Scroll down to turn on the Compare toggle.

Andy says GA4’s date comparison feature is better than the old Universal Analytics because its comparison matches the days of the week between the two periods.

The resulting report identifies rising and falling search traffic as illustrated in this report:

In the report above, the URL — /blog/how-to-write-testimonials-examples — saw a 43.43% drop in views when comparing Jan. 7 through Feb. 3, 2023, to Dec. 10, 2022, through Jan. 6, 2023. But the URL — /blog/website-navigation — saw a 42.86% rise in search traffic during that same period.

Once you’ve identified content with falling search traffic, consider improving it with these ideas:

What phrases does this article rank for?

Knowing the phrases an article ranks for helps you understand why people find the page. It also helps determine how to optimize for additional phrases and drive more organic search traffic to it.

To do this, Andy enlists the help of ChatGPT. However, the first step is to generate a report from Google Search Console:

• Search type: Web
• Date: [a timeframe] (In this example, Andy selected the last 3 months.)
• Query: -orbit (i.e., to exclude branded queries for Andy’s agency Orbit Media)
• Page: [the page URL]

Click “Export” in the upper right to generate a CSV file containing the data.

Feed the CSV file into ChatGPT. Andy uses ChatGPT Plus ($20 a month) because it supports data analysis and file uploads. He starts with this prompt:

“I’m giving you Google Search Console data showing the search performance of a URL. Can you analyze?”

ChatGPT replies with the five columns of data and suggested areas it could analyze (e.g., queries generating the most clicks).

Since any ranking page also ranks for a bunch of other things, he consolidates the list with this prompt:

“You are an expert SEO who is highly proficient at keyword analysis. Merge the rows with similar queries into single rows with a single, representative marketing keyphrase. As you do so, for each row, combine the data for impressions and clicks and averaging the data for CTR and Avg Position. Provide link to download.”

Any time you ask AI to manipulate data, ask it to provide a link to download the data so you can confirm that it processed the data properly. In this instance, Andy finds the data from ChatGPT looks good.

The chart includes 19 rows of queries related to what should be on a website homepage, accompanied by their total clicks, impressions, CTR, and position.

With the data on what phrases the page ranks for, Andy pastes the content into ChatGPT and makes this prompt:

“Here’s the page for the data you just analyzed. Evaluate this content’s keyword usage and frequency based on the keywords in the dataset.”

ChatGPT replies with a set of steps it will perform, including text preprocessing and keyword extraction:

• Text preprocessing: We’ll clean the webpage to prepare it for analysis.
• Keyphrase extraction: We’ll identify the keyphrases that appear in the content.
• Frequency count: We’ll count how often each keyphrase appears in the content.
• Evaluation: Finally, we’ll assess the usage of keyphrases based on their frequency in the content and their importance in the aggregated Google Search Console data.

Andy refines the prompt to have ChatGPT get more sophisticated in its recommendations (bolded emphasis by Andy):

“Suggest edits to this page that would rank higher by better indicating its relevance for the keyphrases in the dataset. Focus on recommendations for including the phrases that are not used on the page but appear in the dataset. Make suggestions that improve the flow and clarity of the content, as well as keyphrase relevance. Highlight the recommended changes.”

What content edits can improve rankings?

I find the results quite amazing. Some of the recommended changes generated by ChatGPT:

Introduction

Current: Social media streams are crowded, fast-flowing, and very competitive. As social networks pull back on organic reach, showing up is hard, and getting attention is even harder.

Suggested edit: Navigating the crowded and competitive social media video landscape can be daunting. With platforms like Facebook and Instagram curtailing organic reach, the challenge of making your DIY social media videos visible is greater than ever.

Second section: Winning the algorithm

Current: Social media videos are so engaging that the networks are pushing them way to the top of social streams. Facebook, Instagram, and LinkedIn all want us to do more in this all-powerful format. It wins in the algorithms and it wins for attention.

Suggested edit: The power of creating engaging social media videos is so compelling that platforms like Facebook, Instagram, and LinkedIn are giving them preferential treatment. This is why learning how to create captivating social media video content is crucial—it wins both in algorithms and in capturing audience attention.

Third section: Tools and equipment

Current: Your laptop has a camera and mic, but you can do better than that. A little more effort will get you a high-quality video and much better results.

Suggested edit: While your laptop’s built-in camera and mic may suffice for starters, investing in specialized video equipment for social media marketing can significantly boost the quality and engagement of your DIY social media videos.

Which articles convert email subscribers the best?

Over the past 18 years, Andy has written over 575 articles on topics ranging from influencer marketing to web analytics to email marketing. Some articles perform amazingly, while others receive no traffic. Similarly, some articles get people to subscribe to Orbit’s email list, while others generate zero subscribers.

“When you know that difference, you know exactly what to do on social media. You know what to promote and what’s working well,” Andy says.

In the past, he used Excel spreadsheets to analyze the ­­ for every article on his site. These days, he uses ChatGPT.

First, he collects the data from GA4. Under the Engagement section, he selects “Page and screens: Page path + query string” report. To customize it, he clicks on the pencil icon in the upper right and adds these metrics:

• Sessions
• Users
• Conversions

TIP: If you don’t see your metric right away, type its name and select it when it appears.

Since Andy only wanted blogs in his data set, he clicks “Add filter” in the upper left and builds the filter with these attributes:

• Dimension: Page path and screen class
• Match type: contains
• Value: blog

From here, click the export icon in the upper right (i.e., it’s two icons to the left of the pencil icon) and select “Download CSV.”

Next, Andy creates a second CSV file by changing “Value: blog” to “Value: /blog-newsletter-thank-you-page” to collect his email signup thank you page data:

He clicks the “+” button left of the report columns to select a secondary dimension of “Page Referrer.” The resulting report shows the pages users read right before they subscribed to Orbit’s email list. He exports this data as a CSV.

Andy cleans up this file by deleting the nine rows of metadata at the top, so ChatGPT won’t get confused in its analysis.

The two files are fed into ChatGPT to confirm it could analyze them. Next, Andy asks ChatGPT to do the equivalent of an Excel VLOOKUP so users can find things in a table or a range by row. He uses this prompt:

“Map the data from the two files together into one file, as you would do with vlookup, so each row is a URL and the sessions and conversions are shown for each.”

He asks ChatGPT to provide a file to download. ChatGPT generates 17 rows of URLs accompanied by sessions (traffic), users (traffic), conversions, sessions, users (converted), and conversions (newsletter):

Then, Andy prompts ChatGPT to draw a bar chart showing the conversion rate for each blog post.

ChatGPT responds with how it calculates the conversion rate for each blog post and visualizes a bar chart showing the top 20 blog posts by conversion rate.

NOTE: Conversion rate data is only available because ChatGPT divided the conversion data (numerator) by session traffic (denominator).

Next steps

To amplify the articles that drive the most email conversions, Andy suggests:

• Promoting the article(s) in your email signature.
• Featuring them on your homepage.
• Putting them into a social media marketing rotation.
• Linking to them in presentations.
• Linking to them from high-traffic pages.

Which email campaigns perform the best?

You can also use the GA4-ChatGPT combination to understand which email campaigns perform the best. Here’s how Andy does it:

• Export GA4 data showing traffic to pages based on UTM codes related to email campaigns.
• Export data from Orbit’s email marketing platform (ActiveCampaign).
• Ask ChatGPT to merge the two datasets, analyze, and provide recommendations

To collect the GA4 data, Andy uses the traffic acquisition report, which can be found under Acquisition in the left navigation bar. He clicks on the pencil in the upper right to customize it and selects the following metrics before clicking the “apply” button:

• Sessions
• Engagement rate
• Average engagement time per session
• Conversions
• Session conversion rate

Next, he clicks “Add filter,” selects the “Session medium” dimension, then the value of “email.”

Andy changes the primary dimension to “Session campaign.” The resulting report has one row for each email campaign (i.e., traffic statistics based on the utm_campaign parameter). Andy exports this data as a CSV file.

Next, Andy exports a data file from ActiveCampaign, which includes:

• Campaign name
• Number of sends
• Number of opens
• Number of clicks
• Number of unsubscribes
• Number of bounces
• Open rate
• Click rate
• Unsubscribe date
• Bounce rate

Andy uploads each file separately to ChatGPT and confirms it can process each file. He then prompts ChatGPT to merge the datasets:

“These two datasets show the performance of the same campaigns from both the email service provider (active campaign) and the website behavior (google analytics). I want to see the performance of each campaign across both datasets. Merge the Google Analytics data with the ActiveCampaign data, combining ‘session campaign’ with ‘campaign name’ into one column, inferring which campaigns were the same based on names and dates. Provide link to download.”

Once ChatGPT merges the datasets, Andy confirms the data looks good and puts ChatGPT to work.

The basic prompt is:

“You are a skilled digital strategist and email marketing expert. What insights can you gather from the combined dataset?”

Moonshot prompts include:

• “What can you tell me about newsletter campaigns?”
• “What insights can you gather from this data?”
• “Make a chart using an unusual visualization method showing unexpected insights from the data in this dataset.”
• “How would Leonardo da Vinci visualize this data? Show your thinking.”

Next, Andy asks ChatGPT to categorize the newsletter campaigns and provide insights on metrics for each category. It returns these strategic insights:

• Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.
• Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.
• SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.
• Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).

Reevaluate SEO campaigns.

Enhance personalization.

Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

You’ve used the GA4 version of Google Analytics for at least a year, if not two to three. You’ve also probably tested the potential of ChatGPT and similar generative AI tools.

But have you combined the power of both?

That’s where Andy Crestodina comes in. The co-founder and CMO of Orbit Media Studios discussed the opportunities in his Marketing Analytics & Data Science (MADS) conference presentation, How to Audit Your Content Using GA4: A Data-Driven Approach to SEO and Lead Generation.

In this article, he walks you through step by step how to create GA4 reports and prompt ChatGPT to identify rising and falling search traffic, improve content rankings, and deliver more effective newsletters.

Which content has rising and falling search traffic?

To see traffic from organic search:

• Go to the Engagement section in the left column.
• In the dropdown, click on the report “Pages and screens: Page title and screen class.”
• Next, click “Add filter” to see organic traffic.

To build the filter:

• Go to the Include dropdown in the right column.
• Select Session medium.
• Under Dimension values, select organic.

At the top of the page, click on “Add comparison.” Select the period, such as the last seven days or the last 28 days. Scroll down to turn on the Compare toggle.

Andy says GA4’s date comparison feature is better than the old Universal Analytics because its comparison matches the days of the week between the two periods.

The resulting report identifies rising and falling search traffic as illustrated in this report:

In the report above, the URL — /blog/how-to-write-testimonials-examples — saw a 43.43% drop in views when comparing Jan. 7 through Feb. 3, 2023, to Dec. 10, 2022, through Jan. 6, 2023. But the URL — /blog/website-navigation — saw a 42.86% rise in search traffic during that same period.

Once you’ve identified content with falling search traffic, consider improving it with these ideas:

What phrases does this article rank for?

Knowing the phrases an article ranks for helps you understand why people find the page. It also helps determine how to optimize for additional phrases and drive more organic search traffic to it.

To do this, Andy enlists the help of ChatGPT. However, the first step is to generate a report from Google Search Console:

• Search type: Web
• Date: [a timeframe] (In this example, Andy selected the last 3 months.)
• Query: -orbit (i.e., to exclude branded queries for Andy’s agency Orbit Media)
• Page: [the page URL]

Click “Export” in the upper right to generate a CSV file containing the data.

Feed the CSV file into ChatGPT. Andy uses ChatGPT Plus ($20 a month) because it supports data analysis and file uploads. He starts with this prompt:

“I’m giving you Google Search Console data showing the search performance of a URL. Can you analyze?”

ChatGPT replies with the five columns of data and suggested areas it could analyze (e.g., queries generating the most clicks).

Since any ranking page also ranks for a bunch of other things, he consolidates the list with this prompt:

“You are an expert SEO who is highly proficient at keyword analysis. Merge the rows with similar queries into single rows with a single, representative marketing keyphrase. As you do so, for each row, combine the data for impressions and clicks and averaging the data for CTR and Avg Position. Provide link to download.”

Any time you ask AI to manipulate data, ask it to provide a link to download the data so you can confirm that it processed the data properly. In this instance, Andy finds the data from ChatGPT looks good.

The chart includes 19 rows of queries related to what should be on a website homepage, accompanied by their total clicks, impressions, CTR, and position.

With the data on what phrases the page ranks for, Andy pastes the content into ChatGPT and makes this prompt:

“Here’s the page for the data you just analyzed. Evaluate this content’s keyword usage and frequency based on the keywords in the dataset.”

ChatGPT replies with a set of steps it will perform, including text preprocessing and keyword extraction:

• Text preprocessing: We’ll clean the webpage to prepare it for analysis.
• Keyphrase extraction: We’ll identify the keyphrases that appear in the content.
• Frequency count: We’ll count how often each keyphrase appears in the content.
• Evaluation: Finally, we’ll assess the usage of keyphrases based on their frequency in the content and their importance in the aggregated Google Search Console data.

Andy refines the prompt to have ChatGPT get more sophisticated in its recommendations (bolded emphasis by Andy):

“Suggest edits to this page that would rank higher by better indicating its relevance for the keyphrases in the dataset. Focus on recommendations for including the phrases that are not used on the page but appear in the dataset. Make suggestions that improve the flow and clarity of the content, as well as keyphrase relevance. Highlight the recommended changes.”

What content edits can improve rankings?

I find the results quite amazing. Some of the recommended changes generated by ChatGPT:

Introduction

Current: Social media streams are crowded, fast-flowing, and very competitive. As social networks pull back on organic reach, showing up is hard, and getting attention is even harder.

Suggested edit: Navigating the crowded and competitive social media video landscape can be daunting. With platforms like Facebook and Instagram curtailing organic reach, the challenge of making your DIY social media videos visible is greater than ever.

Second section: Winning the algorithm

Current: Social media videos are so engaging that the networks are pushing them way to the top of social streams. Facebook, Instagram, and LinkedIn all want us to do more in this all-powerful format. It wins in the algorithms and it wins for attention.

Suggested edit: The power of creating engaging social media videos is so compelling that platforms like Facebook, Instagram, and LinkedIn are giving them preferential treatment. This is why learning how to create captivating social media video content is crucial—it wins both in algorithms and in capturing audience attention.

Third section: Tools and equipment

Current: Your laptop has a camera and mic, but you can do better than that. A little more effort will get you a high-quality video and much better results.

Suggested edit: While your laptop’s built-in camera and mic may suffice for starters, investing in specialized video equipment for social media marketing can significantly boost the quality and engagement of your DIY social media videos.

Which articles convert email subscribers the best?

Over the past 18 years, Andy has written over 575 articles on topics ranging from influencer marketing to web analytics to email marketing. Some articles perform amazingly, while others receive no traffic. Similarly, some articles get people to subscribe to Orbit’s email list, while others generate zero subscribers.

“When you know that difference, you know exactly what to do on social media. You know what to promote and what’s working well,” Andy says.

In the past, he used Excel spreadsheets to analyze the ­­ for every article on his site. These days, he uses ChatGPT.

First, he collects the data from GA4. Under the Engagement section, he selects “Page and screens: Page path + query string” report. To customize it, he clicks on the pencil icon in the upper right and adds these metrics:

• Sessions
• Users
• Conversions

TIP: If you don’t see your metric right away, type its name and select it when it appears.

Since Andy only wanted blogs in his data set, he clicks “Add filter” in the upper left and builds the filter with these attributes:

• Dimension: Page path and screen class
• Match type: contains
• Value: blog

From here, click the export icon in the upper right (i.e., it’s two icons to the left of the pencil icon) and select “Download CSV.”

Next, Andy creates a second CSV file by changing “Value: blog” to “Value: /blog-newsletter-thank-you-page” to collect his email signup thank you page data:

He clicks the “+” button left of the report columns to select a secondary dimension of “Page Referrer.” The resulting report shows the pages users read right before they subscribed to Orbit’s email list. He exports this data as a CSV.

Andy cleans up this file by deleting the nine rows of metadata at the top, so ChatGPT won’t get confused in its analysis.

The two files are fed into ChatGPT to confirm it could analyze them. Next, Andy asks ChatGPT to do the equivalent of an Excel VLOOKUP so users can find things in a table or a range by row. He uses this prompt:

“Map the data from the two files together into one file, as you would do with vlookup, so each row is a URL and the sessions and conversions are shown for each.”

He asks ChatGPT to provide a file to download. ChatGPT generates 17 rows of URLs accompanied by sessions (traffic), users (traffic), conversions, sessions, users (converted), and conversions (newsletter):

Then, Andy prompts ChatGPT to draw a bar chart showing the conversion rate for each blog post.

ChatGPT responds with how it calculates the conversion rate for each blog post and visualizes a bar chart showing the top 20 blog posts by conversion rate.

NOTE: Conversion rate data is only available because ChatGPT divided the conversion data (numerator) by session traffic (denominator).

Next steps

To amplify the articles that drive the most email conversions, Andy suggests:

• Promoting the article(s) in your email signature.
• Featuring them on your homepage.
• Putting them into a social media marketing rotation.
• Linking to them in presentations.
• Linking to them from high-traffic pages.

Which email campaigns perform the best?

You can also use the GA4-ChatGPT combination to understand which email campaigns perform the best. Here’s how Andy does it:

• Export GA4 data showing traffic to pages based on UTM codes related to email campaigns.
• Export data from Orbit’s email marketing platform (ActiveCampaign).
• Ask ChatGPT to merge the two datasets, analyze, and provide recommendations

To collect the GA4 data, Andy uses the traffic acquisition report, which can be found under Acquisition in the left navigation bar. He clicks on the pencil in the upper right to customize it and selects the following metrics before clicking the “apply” button:

• Sessions
• Engagement rate
• Average engagement time per session
• Conversions
• Session conversion rate

Next, he clicks “Add filter,” selects the “Session medium” dimension, then the value of “email.”

Andy changes the primary dimension to “Session campaign.” The resulting report has one row for each email campaign (i.e., traffic statistics based on the utm_campaign parameter). Andy exports this data as a CSV file.

Next, Andy exports a data file from ActiveCampaign, which includes:

• Campaign name
• Number of sends
• Number of opens
• Number of clicks
• Number of unsubscribes
• Number of bounces
• Open rate
• Click rate
• Unsubscribe date
• Bounce rate

Andy uploads each file separately to ChatGPT and confirms it can process each file. He then prompts ChatGPT to merge the datasets:

“These two datasets show the performance of the same campaigns from both the email service provider (active campaign) and the website behavior (google analytics). I want to see the performance of each campaign across both datasets. Merge the Google Analytics data with the ActiveCampaign data, combining ‘session campaign’ with ‘campaign name’ into one column, inferring which campaigns were the same based on names and dates. Provide link to download.”

Once ChatGPT merges the datasets, Andy confirms the data looks good and puts ChatGPT to work.

The basic prompt is:

“You are a skilled digital strategist and email marketing expert. What insights can you gather from the combined dataset?”

Moonshot prompts include:

• “What can you tell me about newsletter campaigns?”
• “What insights can you gather from this data?”
• “Make a chart using an unusual visualization method showing unexpected insights from the data in this dataset.”
• “How would Leonardo da Vinci visualize this data? Show your thinking.”

Next, Andy asks ChatGPT to categorize the newsletter campaigns and provide insights on metrics for each category. It returns these strategic insights:

• Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.
• Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.
• SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.
• Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.

Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.

SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.

Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

You’ve used the GA4 version of Google Analytics for at least a year, if not two to three. You’ve also probably tested the potential of ChatGPT and similar generative AI tools.

But have you combined the power of both?

That’s where Andy Crestodina comes in. The co-founder and CMO of Orbit Media Studios discussed the opportunities in his Marketing Analytics & Data Science (MADS) conference presentation, How to Audit Your Content Using GA4: A Data-Driven Approach to SEO and Lead Generation.

In this article, he walks you through step by step how to create GA4 reports and prompt ChatGPT to identify rising and falling search traffic, improve content rankings, and deliver more effective newsletters.

Which content has rising and falling search traffic?

To see traffic from organic search:

• Go to the Engagement section in the left column.
• In the dropdown, click on the report “Pages and screens: Page title and screen class.”
• Next, click “Add filter” to see organic traffic.

To build the filter:

• Go to the Include dropdown in the right column.
• Select Session medium.
• Under Dimension values, select organic.

At the top of the page, click on “Add comparison.” Select the period, such as the last seven days or the last 28 days. Scroll down to turn on the Compare toggle.

Andy says GA4’s date comparison feature is better than the old Universal Analytics because its comparison matches the days of the week between the two periods.

The resulting report identifies rising and falling search traffic as illustrated in this report:

In the report above, the URL — /blog/how-to-write-testimonials-examples — saw a 43.43% drop in views when comparing Jan. 7 through Feb. 3, 2023, to Dec. 10, 2022, through Jan. 6, 2023. But the URL — /blog/website-navigation — saw a 42.86% rise in search traffic during that same period.

Once you’ve identified content with falling search traffic, consider improving it with these ideas:

What phrases does this article rank for?

Knowing the phrases an article ranks for helps you understand why people find the page. It also helps determine how to optimize for additional phrases and drive more organic search traffic to it.

To do this, Andy enlists the help of ChatGPT. However, the first step is to generate a report from Google Search Console:

• Search type: Web
• Date: [a timeframe] (In this example, Andy selected the last 3 months.)
• Query: -orbit (i.e., to exclude branded queries for Andy’s agency Orbit Media)
• Page: [the page URL]

Click “Export” in the upper right to generate a CSV file containing the data.

Feed the CSV file into ChatGPT. Andy uses ChatGPT Plus ($20 a month) because it supports data analysis and file uploads. He starts with this prompt:

“I’m giving you Google Search Console data showing the search performance of a URL. Can you analyze?”

ChatGPT replies with the five columns of data and suggested areas it could analyze (e.g., queries generating the most clicks).

Since any ranking page also ranks for a bunch of other things, he consolidates the list with this prompt:

“You are an expert SEO who is highly proficient at keyword analysis. Merge the rows with similar queries into single rows with a single, representative marketing keyphrase. As you do so, for each row, combine the data for impressions and clicks and averaging the data for CTR and Avg Position. Provide link to download.”

Any time you ask AI to manipulate data, ask it to provide a link to download the data so you can confirm that it processed the data properly. In this instance, Andy finds the data from ChatGPT looks good.

The chart includes 19 rows of queries related to what should be on a website homepage, accompanied by their total clicks, impressions, CTR, and position.

With the data on what phrases the page ranks for, Andy pastes the content into ChatGPT and makes this prompt:

“Here’s the page for the data you just analyzed. Evaluate this content’s keyword usage and frequency based on the keywords in the dataset.”

ChatGPT replies with a set of steps it will perform, including text preprocessing and keyword extraction:

• Text preprocessing: We’ll clean the webpage to prepare it for analysis.
• Keyphrase extraction: We’ll identify the keyphrases that appear in the content.
• Frequency count: We’ll count how often each keyphrase appears in the content.
• Evaluation: Finally, we’ll assess the usage of keyphrases based on their frequency in the content and their importance in the aggregated Google Search Console data.

Andy refines the prompt to have ChatGPT get more sophisticated in its recommendations (bolded emphasis by Andy):

“Suggest edits to this page that would rank higher by better indicating its relevance for the keyphrases in the dataset. Focus on recommendations for including the phrases that are not used on the page but appear in the dataset. Make suggestions that improve the flow and clarity of the content, as well as keyphrase relevance. Highlight the recommended changes.”

What content edits can improve rankings?

I find the results quite amazing. Some of the recommended changes generated by ChatGPT:

Introduction

Current: Social media streams are crowded, fast-flowing, and very competitive. As social networks pull back on organic reach, showing up is hard, and getting attention is even harder.

Suggested edit: Navigating the crowded and competitive social media video landscape can be daunting. With platforms like Facebook and Instagram curtailing organic reach, the challenge of making your DIY social media videos visible is greater than ever.

Second section: Winning the algorithm

Current: Social media videos are so engaging that the networks are pushing them way to the top of social streams. Facebook, Instagram, and LinkedIn all want us to do more in this all-powerful format. It wins in the algorithms and it wins for attention.

Suggested edit: The power of creating engaging social media videos is so compelling that platforms like Facebook, Instagram, and LinkedIn are giving them preferential treatment. This is why learning how to create captivating social media video content is crucial—it wins both in algorithms and in capturing audience attention.

Third section: Tools and equipment

Current: Your laptop has a camera and mic, but you can do better than that. A little more effort will get you a high-quality video and much better results.

Suggested edit: While your laptop’s built-in camera and mic may suffice for starters, investing in specialized video equipment for social media marketing can significantly boost the quality and engagement of your DIY social media videos.

Which articles convert email subscribers the best?

Over the past 18 years, Andy has written over 575 articles on topics ranging from influencer marketing to web analytics to email marketing. Some articles perform amazingly, while others receive no traffic. Similarly, some articles get people to subscribe to Orbit’s email list, while others generate zero subscribers.

“When you know that difference, you know exactly what to do on social media. You know what to promote and what’s working well,” Andy says.

In the past, he used Excel spreadsheets to analyze the ­­ for every article on his site. These days, he uses ChatGPT.

First, he collects the data from GA4. Under the Engagement section, he selects “Page and screens: Page path + query string” report. To customize it, he clicks on the pencil icon in the upper right and adds these metrics:

• Sessions
• Users
• Conversions

TIP: If you don’t see your metric right away, type its name and select it when it appears.

Since Andy only wanted blogs in his data set, he clicks “Add filter” in the upper left and builds the filter with these attributes:

• Dimension: Page path and screen class
• Match type: contains
• Value: blog

From here, click the export icon in the upper right (i.e., it’s two icons to the left of the pencil icon) and select “Download CSV.”

Next, Andy creates a second CSV file by changing “Value: blog” to “Value: /blog-newsletter-thank-you-page” to collect his email signup thank you page data:

He clicks the “+” button left of the report columns to select a secondary dimension of “Page Referrer.” The resulting report shows the pages users read right before they subscribed to Orbit’s email list. He exports this data as a CSV.

Andy cleans up this file by deleting the nine rows of metadata at the top, so ChatGPT won’t get confused in its analysis.

The two files are fed into ChatGPT to confirm it could analyze them. Next, Andy asks ChatGPT to do the equivalent of an Excel VLOOKUP so users can find things in a table or a range by row. He uses this prompt:

“Map the data from the two files together into one file, as you would do with vlookup, so each row is a URL and the sessions and conversions are shown for each.”

He asks ChatGPT to provide a file to download. ChatGPT generates 17 rows of URLs accompanied by sessions (traffic), users (traffic), conversions, sessions, users (converted), and conversions (newsletter):

Then, Andy prompts ChatGPT to draw a bar chart showing the conversion rate for each blog post.

ChatGPT responds with how it calculates the conversion rate for each blog post and visualizes a bar chart showing the top 20 blog posts by conversion rate.

NOTE: Conversion rate data is only available because ChatGPT divided the conversion data (numerator) by session traffic (denominator).

Next steps

To amplify the articles that drive the most email conversions, Andy suggests:

• Promoting the article(s) in your email signature.
• Featuring them on your homepage.
• Putting them into a social media marketing rotation.
• Linking to them in presentations.
• Linking to them from high-traffic pages.

Which email campaigns perform the best?

You can also use the GA4-ChatGPT combination to understand which email campaigns perform the best. Here’s how Andy does it:

• Export GA4 data showing traffic to pages based on UTM codes related to email campaigns.
• Export data from Orbit’s email marketing platform (ActiveCampaign).
• Ask ChatGPT to merge the two datasets, analyze, and provide recommendations

To collect the GA4 data, Andy uses the traffic acquisition report, which can be found under Acquisition in the left navigation bar. He clicks on the pencil in the upper right to customize it and selects the following metrics before clicking the “apply” button:

• Sessions
• Engagement rate
• Average engagement time per session
• Conversions
• Session conversion rate

Next, he clicks “Add filter,” selects the “Session medium” dimension, then the value of “email.”

Andy changes the primary dimension to “Session campaign.” The resulting report has one row for each email campaign (i.e., traffic statistics based on the utm_campaign parameter). Andy exports this data as a CSV file.

Next, Andy exports a data file from ActiveCampaign, which includes:

• Campaign name
• Number of sends
• Number of opens
• Number of clicks
• Number of unsubscribes
• Number of bounces
• Open rate
• Click rate
• Unsubscribe date
• Bounce rate

Andy uploads each file separately to ChatGPT and confirms it can process each file. He then prompts ChatGPT to merge the datasets:

“These two datasets show the performance of the same campaigns from both the email service provider (active campaign) and the website behavior (google analytics). I want to see the performance of each campaign across both datasets. Merge the Google Analytics data with the ActiveCampaign data, combining ‘session campaign’ with ‘campaign name’ into one column, inferring which campaigns were the same based on names and dates. Provide link to download.”

Once ChatGPT merges the datasets, Andy confirms the data looks good and puts ChatGPT to work.

The basic prompt is:

“You are a skilled digital strategist and email marketing expert. What insights can you gather from the combined dataset?”

Moonshot prompts include:

• “What can you tell me about newsletter campaigns?”
• “What insights can you gather from this data?”
• “Make a chart using an unusual visualization method showing unexpected insights from the data in this dataset.”
• “How would Leonardo da Vinci visualize this data? Show your thinking.”

Next, Andy asks ChatGPT to categorize the newsletter campaigns and provide insights on metrics for each category. It returns these strategic insights:

• Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.
• Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.
• SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.
• Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Report: Acquisition > Traffic acquisition: Search default channel group

Dimension: Session campaign

Secondary dimension: Date

Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.

Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.

SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.

Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

You’ve used the GA4 version of Google Analytics for at least a year, if not two to three. You’ve also probably tested the potential of ChatGPT and similar generative AI tools.

But have you combined the power of both?

That’s where Andy Crestodina comes in. The co-founder and CMO of Orbit Media Studios discussed the opportunities in his Marketing Analytics & Data Science (MADS) conference presentation, How to Audit Your Content Using GA4: A Data-Driven Approach to SEO and Lead Generation.

In this article, he walks you through step by step how to create GA4 reports and prompt ChatGPT to identify rising and falling search traffic, improve content rankings, and deliver more effective newsletters.

Which content has rising and falling search traffic?

To see traffic from organic search:

• Go to the Engagement section in the left column.
• In the dropdown, click on the report “Pages and screens: Page title and screen class.”
• Next, click “Add filter” to see organic traffic.

To build the filter:

• Go to the Include dropdown in the right column.
• Select Session medium.
• Under Dimension values, select organic.

At the top of the page, click on “Add comparison.” Select the period, such as the last seven days or the last 28 days. Scroll down to turn on the Compare toggle.

Andy says GA4’s date comparison feature is better than the old Universal Analytics because its comparison matches the days of the week between the two periods.

The resulting report identifies rising and falling search traffic as illustrated in this report:

In the report above, the URL — /blog/how-to-write-testimonials-examples — saw a 43.43% drop in views when comparing Jan. 7 through Feb. 3, 2023, to Dec. 10, 2022, through Jan. 6, 2023. But the URL — /blog/website-navigation — saw a 42.86% rise in search traffic during that same period.

Once you’ve identified content with falling search traffic, consider improving it with these ideas:

What phrases does this article rank for?

Knowing the phrases an article ranks for helps you understand why people find the page. It also helps determine how to optimize for additional phrases and drive more organic search traffic to it.

To do this, Andy enlists the help of ChatGPT. However, the first step is to generate a report from Google Search Console:

• Search type: Web
• Date: [a timeframe] (In this example, Andy selected the last 3 months.)
• Query: -orbit (i.e., to exclude branded queries for Andy’s agency Orbit Media)
• Page: [the page URL]

Click “Export” in the upper right to generate a CSV file containing the data.

Feed the CSV file into ChatGPT. Andy uses ChatGPT Plus ($20 a month) because it supports data analysis and file uploads. He starts with this prompt:

“I’m giving you Google Search Console data showing the search performance of a URL. Can you analyze?”

ChatGPT replies with the five columns of data and suggested areas it could analyze (e.g., queries generating the most clicks).

Since any ranking page also ranks for a bunch of other things, he consolidates the list with this prompt:

“You are an expert SEO who is highly proficient at keyword analysis. Merge the rows with similar queries into single rows with a single, representative marketing keyphrase. As you do so, for each row, combine the data for impressions and clicks and averaging the data for CTR and Avg Position. Provide link to download.”

Any time you ask AI to manipulate data, ask it to provide a link to download the data so you can confirm that it processed the data properly. In this instance, Andy finds the data from ChatGPT looks good.

The chart includes 19 rows of queries related to what should be on a website homepage, accompanied by their total clicks, impressions, CTR, and position.

With the data on what phrases the page ranks for, Andy pastes the content into ChatGPT and makes this prompt:

“Here’s the page for the data you just analyzed. Evaluate this content’s keyword usage and frequency based on the keywords in the dataset.”

ChatGPT replies with a set of steps it will perform, including text preprocessing and keyword extraction:

• Text preprocessing: We’ll clean the webpage to prepare it for analysis.
• Keyphrase extraction: We’ll identify the keyphrases that appear in the content.
• Frequency count: We’ll count how often each keyphrase appears in the content.
• Evaluation: Finally, we’ll assess the usage of keyphrases based on their frequency in the content and their importance in the aggregated Google Search Console data.

Andy refines the prompt to have ChatGPT get more sophisticated in its recommendations (bolded emphasis by Andy):

“Suggest edits to this page that would rank higher by better indicating its relevance for the keyphrases in the dataset. Focus on recommendations for including the phrases that are not used on the page but appear in the dataset. Make suggestions that improve the flow and clarity of the content, as well as keyphrase relevance. Highlight the recommended changes.”

What content edits can improve rankings?

I find the results quite amazing. Some of the recommended changes generated by ChatGPT:

Introduction

Current: Social media streams are crowded, fast-flowing, and very competitive. As social networks pull back on organic reach, showing up is hard, and getting attention is even harder.

Suggested edit: Navigating the crowded and competitive social media video landscape can be daunting. With platforms like Facebook and Instagram curtailing organic reach, the challenge of making your DIY social media videos visible is greater than ever.

Second section: Winning the algorithm

Current: Social media videos are so engaging that the networks are pushing them way to the top of social streams. Facebook, Instagram, and LinkedIn all want us to do more in this all-powerful format. It wins in the algorithms and it wins for attention.

Suggested edit: The power of creating engaging social media videos is so compelling that platforms like Facebook, Instagram, and LinkedIn are giving them preferential treatment. This is why learning how to create captivating social media video content is crucial—it wins both in algorithms and in capturing audience attention.

Third section: Tools and equipment

Current: Your laptop has a camera and mic, but you can do better than that. A little more effort will get you a high-quality video and much better results.

Suggested edit: While your laptop’s built-in camera and mic may suffice for starters, investing in specialized video equipment for social media marketing can significantly boost the quality and engagement of your DIY social media videos.

Which articles convert email subscribers the best?

Over the past 18 years, Andy has written over 575 articles on topics ranging from influencer marketing to web analytics to email marketing. Some articles perform amazingly, while others receive no traffic. Similarly, some articles get people to subscribe to Orbit’s email list, while others generate zero subscribers.

“When you know that difference, you know exactly what to do on social media. You know what to promote and what’s working well,” Andy says.

In the past, he used Excel spreadsheets to analyze the ­­ for every article on his site. These days, he uses ChatGPT.

First, he collects the data from GA4. Under the Engagement section, he selects “Page and screens: Page path + query string” report. To customize it, he clicks on the pencil icon in the upper right and adds these metrics:

• Sessions
• Users
• Conversions

TIP: If you don’t see your metric right away, type its name and select it when it appears.

Since Andy only wanted blogs in his data set, he clicks “Add filter” in the upper left and builds the filter with these attributes:

• Dimension: Page path and screen class
• Match type: contains
• Value: blog

From here, click the export icon in the upper right (i.e., it’s two icons to the left of the pencil icon) and select “Download CSV.”

Next, Andy creates a second CSV file by changing “Value: blog” to “Value: /blog-newsletter-thank-you-page” to collect his email signup thank you page data:

He clicks the “+” button left of the report columns to select a secondary dimension of “Page Referrer.” The resulting report shows the pages users read right before they subscribed to Orbit’s email list. He exports this data as a CSV.

Andy cleans up this file by deleting the nine rows of metadata at the top, so ChatGPT won’t get confused in its analysis.

The two files are fed into ChatGPT to confirm it could analyze them. Next, Andy asks ChatGPT to do the equivalent of an Excel VLOOKUP so users can find things in a table or a range by row. He uses this prompt:

“Map the data from the two files together into one file, as you would do with vlookup, so each row is a URL and the sessions and conversions are shown for each.”

He asks ChatGPT to provide a file to download. ChatGPT generates 17 rows of URLs accompanied by sessions (traffic), users (traffic), conversions, sessions, users (converted), and conversions (newsletter):

Then, Andy prompts ChatGPT to draw a bar chart showing the conversion rate for each blog post.

ChatGPT responds with how it calculates the conversion rate for each blog post and visualizes a bar chart showing the top 20 blog posts by conversion rate.

NOTE: Conversion rate data is only available because ChatGPT divided the conversion data (numerator) by session traffic (denominator).

Next steps

To amplify the articles that drive the most email conversions, Andy suggests:

• Promoting the article(s) in your email signature.
• Featuring them on your homepage.
• Putting them into a social media marketing rotation.
• Linking to them in presentations.
• Linking to them from high-traffic pages.

Which email campaigns perform the best?

You can also use the GA4-ChatGPT combination to understand which email campaigns perform the best. Here’s how Andy does it:

• Export GA4 data showing traffic to pages based on UTM codes related to email campaigns.
• Export data from Orbit’s email marketing platform (ActiveCampaign).
• Ask ChatGPT to merge the two datasets, analyze, and provide recommendations

To collect the GA4 data, Andy uses the traffic acquisition report, which can be found under Acquisition in the left navigation bar. He clicks on the pencil in the upper right to customize it and selects the following metrics before clicking the “apply” button:

• Sessions
• Engagement rate
• Average engagement time per session
• Conversions
• Session conversion rate

Next, he clicks “Add filter,” selects the “Session medium” dimension, then the value of “email.”

Andy changes the primary dimension to “Session campaign.” The resulting report has one row for each email campaign (i.e., traffic statistics based on the utm_campaign parameter). Andy exports this data as a CSV file.

Next, Andy exports a data file from ActiveCampaign, which includes:

• Campaign name
• Number of sends
• Number of opens
• Number of clicks
• Number of unsubscribes
• Number of bounces
• Open rate
• Click rate
• Unsubscribe date
• Bounce rate

Andy uploads each file separately to ChatGPT and confirms it can process each file. He then prompts ChatGPT to merge the datasets:

“These two datasets show the performance of the same campaigns from both the email service provider (active campaign) and the website behavior (google analytics). I want to see the performance of each campaign across both datasets. Merge the Google Analytics data with the ActiveCampaign data, combining ‘session campaign’ with ‘campaign name’ into one column, inferring which campaigns were the same based on names and dates. Provide link to download.”

Once ChatGPT merges the datasets, Andy confirms the data looks good and puts ChatGPT to work.

The basic prompt is:

“You are a skilled digital strategist and email marketing expert. What insights can you gather from the combined dataset?”

Moonshot prompts include:

• “What can you tell me about newsletter campaigns?”
• “What insights can you gather from this data?”
• “Make a chart using an unusual visualization method showing unexpected insights from the data in this dataset.”
• “How would Leonardo da Vinci visualize this data? Show your thinking.”

Next, Andy asks ChatGPT to categorize the newsletter campaigns and provide insights on metrics for each category. It returns these strategic insights:

• Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.
• Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.
• SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.
• Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

“Based on this data, what type of campaigns should be sent more often?”

“What types of campaigns should be abandoned?”

“Suggest changes that would improve the efficiency of this email program.”

“Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.

Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.

SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.

Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

You’ve used the GA4 version of Google Analytics for at least a year, if not two to three. You’ve also probably tested the potential of ChatGPT and similar generative AI tools.

But have you combined the power of both?

That’s where Andy Crestodina comes in. The co-founder and CMO of Orbit Media Studios discussed the opportunities in his Marketing Analytics & Data Science (MADS) conference presentation, How to Audit Your Content Using GA4: A Data-Driven Approach to SEO and Lead Generation.

In this article, he walks you through step by step how to create GA4 reports and prompt ChatGPT to identify rising and falling search traffic, improve content rankings, and deliver more effective newsletters.

Which content has rising and falling search traffic?

To see traffic from organic search:

• Go to the Engagement section in the left column.
• In the dropdown, click on the report “Pages and screens: Page title and screen class.”
• Next, click “Add filter” to see organic traffic.

To build the filter:

• Go to the Include dropdown in the right column.
• Select Session medium.
• Under Dimension values, select organic.

At the top of the page, click on “Add comparison.” Select the period, such as the last seven days or the last 28 days. Scroll down to turn on the Compare toggle.

Andy says GA4’s date comparison feature is better than the old Universal Analytics because its comparison matches the days of the week between the two periods.

The resulting report identifies rising and falling search traffic as illustrated in this report:

In the report above, the URL — /blog/how-to-write-testimonials-examples — saw a 43.43% drop in views when comparing Jan. 7 through Feb. 3, 2023, to Dec. 10, 2022, through Jan. 6, 2023. But the URL — /blog/website-navigation — saw a 42.86% rise in search traffic during that same period.

Once you’ve identified content with falling search traffic, consider improving it with these ideas:

What phrases does this article rank for?

Knowing the phrases an article ranks for helps you understand why people find the page. It also helps determine how to optimize for additional phrases and drive more organic search traffic to it.

To do this, Andy enlists the help of ChatGPT. However, the first step is to generate a report from Google Search Console:

• Search type: Web
• Date: [a timeframe] (In this example, Andy selected the last 3 months.)
• Query: -orbit (i.e., to exclude branded queries for Andy’s agency Orbit Media)
• Page: [the page URL]

Click “Export” in the upper right to generate a CSV file containing the data.

Feed the CSV file into ChatGPT. Andy uses ChatGPT Plus ($20 a month) because it supports data analysis and file uploads. He starts with this prompt:

“I’m giving you Google Search Console data showing the search performance of a URL. Can you analyze?”

ChatGPT replies with the five columns of data and suggested areas it could analyze (e.g., queries generating the most clicks).

Since any ranking page also ranks for a bunch of other things, he consolidates the list with this prompt:

“You are an expert SEO who is highly proficient at keyword analysis. Merge the rows with similar queries into single rows with a single, representative marketing keyphrase. As you do so, for each row, combine the data for impressions and clicks and averaging the data for CTR and Avg Position. Provide link to download.”

Any time you ask AI to manipulate data, ask it to provide a link to download the data so you can confirm that it processed the data properly. In this instance, Andy finds the data from ChatGPT looks good.

The chart includes 19 rows of queries related to what should be on a website homepage, accompanied by their total clicks, impressions, CTR, and position.

With the data on what phrases the page ranks for, Andy pastes the content into ChatGPT and makes this prompt:

“Here’s the page for the data you just analyzed. Evaluate this content’s keyword usage and frequency based on the keywords in the dataset.”

ChatGPT replies with a set of steps it will perform, including text preprocessing and keyword extraction:

• Text preprocessing: We’ll clean the webpage to prepare it for analysis.
• Keyphrase extraction: We’ll identify the keyphrases that appear in the content.
• Frequency count: We’ll count how often each keyphrase appears in the content.
• Evaluation: Finally, we’ll assess the usage of keyphrases based on their frequency in the content and their importance in the aggregated Google Search Console data.

Andy refines the prompt to have ChatGPT get more sophisticated in its recommendations (bolded emphasis by Andy):

“Suggest edits to this page that would rank higher by better indicating its relevance for the keyphrases in the dataset. Focus on recommendations for including the phrases that are not used on the page but appear in the dataset. Make suggestions that improve the flow and clarity of the content, as well as keyphrase relevance. Highlight the recommended changes.”

What content edits can improve rankings?

I find the results quite amazing. Some of the recommended changes generated by ChatGPT:

Introduction

Current: Social media streams are crowded, fast-flowing, and very competitive. As social networks pull back on organic reach, showing up is hard, and getting attention is even harder.

Suggested edit: Navigating the crowded and competitive social media video landscape can be daunting. With platforms like Facebook and Instagram curtailing organic reach, the challenge of making your DIY social media videos visible is greater than ever.

Second section: Winning the algorithm

Current: Social media videos are so engaging that the networks are pushing them way to the top of social streams. Facebook, Instagram, and LinkedIn all want us to do more in this all-powerful format. It wins in the algorithms and it wins for attention.

Suggested edit: The power of creating engaging social media videos is so compelling that platforms like Facebook, Instagram, and LinkedIn are giving them preferential treatment. This is why learning how to create captivating social media video content is crucial—it wins both in algorithms and in capturing audience attention.

Third section: Tools and equipment

Current: Your laptop has a camera and mic, but you can do better than that. A little more effort will get you a high-quality video and much better results.

Suggested edit: While your laptop’s built-in camera and mic may suffice for starters, investing in specialized video equipment for social media marketing can significantly boost the quality and engagement of your DIY social media videos.

Which articles convert email subscribers the best?

Over the past 18 years, Andy has written over 575 articles on topics ranging from influencer marketing to web analytics to email marketing. Some articles perform amazingly, while others receive no traffic. Similarly, some articles get people to subscribe to Orbit’s email list, while others generate zero subscribers.

“When you know that difference, you know exactly what to do on social media. You know what to promote and what’s working well,” Andy says.

In the past, he used Excel spreadsheets to analyze the ­­ for every article on his site. These days, he uses ChatGPT.

First, he collects the data from GA4. Under the Engagement section, he selects “Page and screens: Page path + query string” report. To customize it, he clicks on the pencil icon in the upper right and adds these metrics:

• Sessions
• Users
• Conversions

TIP: If you don’t see your metric right away, type its name and select it when it appears.

Since Andy only wanted blogs in his data set, he clicks “Add filter” in the upper left and builds the filter with these attributes:

• Dimension: Page path and screen class
• Match type: contains
• Value: blog

From here, click the export icon in the upper right (i.e., it’s two icons to the left of the pencil icon) and select “Download CSV.”

Next, Andy creates a second CSV file by changing “Value: blog” to “Value: /blog-newsletter-thank-you-page” to collect his email signup thank you page data:

He clicks the “+” button left of the report columns to select a secondary dimension of “Page Referrer.” The resulting report shows the pages users read right before they subscribed to Orbit’s email list. He exports this data as a CSV.

Andy cleans up this file by deleting the nine rows of metadata at the top, so ChatGPT won’t get confused in its analysis.

The two files are fed into ChatGPT to confirm it could analyze them. Next, Andy asks ChatGPT to do the equivalent of an Excel VLOOKUP so users can find things in a table or a range by row. He uses this prompt:

“Map the data from the two files together into one file, as you would do with vlookup, so each row is a URL and the sessions and conversions are shown for each.”

He asks ChatGPT to provide a file to download. ChatGPT generates 17 rows of URLs accompanied by sessions (traffic), users (traffic), conversions, sessions, users (converted), and conversions (newsletter):

Then, Andy prompts ChatGPT to draw a bar chart showing the conversion rate for each blog post.

ChatGPT responds with how it calculates the conversion rate for each blog post and visualizes a bar chart showing the top 20 blog posts by conversion rate.

NOTE: Conversion rate data is only available because ChatGPT divided the conversion data (numerator) by session traffic (denominator).

Next steps

To amplify the articles that drive the most email conversions, Andy suggests:

• Promoting the article(s) in your email signature.
• Featuring them on your homepage.
• Putting them into a social media marketing rotation.
• Linking to them in presentations.
• Linking to them from high-traffic pages.

Which email campaigns perform the best?

You can also use the GA4-ChatGPT combination to understand which email campaigns perform the best. Here’s how Andy does it:

• Export GA4 data showing traffic to pages based on UTM codes related to email campaigns.
• Export data from Orbit’s email marketing platform (ActiveCampaign).
• Ask ChatGPT to merge the two datasets, analyze, and provide recommendations

To collect the GA4 data, Andy uses the traffic acquisition report, which can be found under Acquisition in the left navigation bar. He clicks on the pencil in the upper right to customize it and selects the following metrics before clicking the “apply” button:

• Sessions
• Engagement rate
• Average engagement time per session
• Conversions
• Session conversion rate

Next, he clicks “Add filter,” selects the “Session medium” dimension, then the value of “email.”

Andy changes the primary dimension to “Session campaign.” The resulting report has one row for each email campaign (i.e., traffic statistics based on the utm_campaign parameter). Andy exports this data as a CSV file.

Next, Andy exports a data file from ActiveCampaign, which includes:

• Campaign name
• Number of sends
• Number of opens
• Number of clicks
• Number of unsubscribes
• Number of bounces
• Open rate
• Click rate
• Unsubscribe date
• Bounce rate

Andy uploads each file separately to ChatGPT and confirms it can process each file. He then prompts ChatGPT to merge the datasets:

“These two datasets show the performance of the same campaigns from both the email service provider (active campaign) and the website behavior (google analytics). I want to see the performance of each campaign across both datasets. Merge the Google Analytics data with the ActiveCampaign data, combining ‘session campaign’ with ‘campaign name’ into one column, inferring which campaigns were the same based on names and dates. Provide link to download.”

Once ChatGPT merges the datasets, Andy confirms the data looks good and puts ChatGPT to work.

The basic prompt is:

“You are a skilled digital strategist and email marketing expert. What insights can you gather from the combined dataset?”

Moonshot prompts include:

• “What can you tell me about newsletter campaigns?”
• “What insights can you gather from this data?”
• “Make a chart using an unusual visualization method showing unexpected insights from the data in this dataset.”
• “How would Leonardo da Vinci visualize this data? Show your thinking.”

Next, Andy asks ChatGPT to categorize the newsletter campaigns and provide insights on metrics for each category. It returns these strategic insights:

• Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.
• Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.
• SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.
• Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).

Reevaluate SEO campaigns.

Enhance personalization.

Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.

Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.

SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.

Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

You’ve used the GA4 version of Google Analytics for at least a year, if not two to three. You’ve also probably tested the potential of ChatGPT and similar generative AI tools.

But have you combined the power of both?

That’s where Andy Crestodina comes in. The co-founder and CMO of Orbit Media Studios discussed the opportunities in his Marketing Analytics & Data Science (MADS) conference presentation, How to Audit Your Content Using GA4: A Data-Driven Approach to SEO and Lead Generation.

In this article, he walks you through step by step how to create GA4 reports and prompt ChatGPT to identify rising and falling search traffic, improve content rankings, and deliver more effective newsletters.

Which content has rising and falling search traffic?

To see traffic from organic search:

• Go to the Engagement section in the left column.
• In the dropdown, click on the report “Pages and screens: Page title and screen class.”
• Next, click “Add filter” to see organic traffic.

To build the filter:

• Go to the Include dropdown in the right column.
• Select Session medium.
• Under Dimension values, select organic.

At the top of the page, click on “Add comparison.” Select the period, such as the last seven days or the last 28 days. Scroll down to turn on the Compare toggle.

Andy says GA4’s date comparison feature is better than the old Universal Analytics because its comparison matches the days of the week between the two periods.

The resulting report identifies rising and falling search traffic as illustrated in this report:

In the report above, the URL — /blog/how-to-write-testimonials-examples — saw a 43.43% drop in views when comparing Jan. 7 through Feb. 3, 2023, to Dec. 10, 2022, through Jan. 6, 2023. But the URL — /blog/website-navigation — saw a 42.86% rise in search traffic during that same period.

Once you’ve identified content with falling search traffic, consider improving it with these ideas:

What phrases does this article rank for?

Knowing the phrases an article ranks for helps you understand why people find the page. It also helps determine how to optimize for additional phrases and drive more organic search traffic to it.

To do this, Andy enlists the help of ChatGPT. However, the first step is to generate a report from Google Search Console:

• Search type: Web
• Date: [a timeframe] (In this example, Andy selected the last 3 months.)
• Query: -orbit (i.e., to exclude branded queries for Andy’s agency Orbit Media)
• Page: [the page URL]

Click “Export” in the upper right to generate a CSV file containing the data.

Feed the CSV file into ChatGPT. Andy uses ChatGPT Plus ($20 a month) because it supports data analysis and file uploads. He starts with this prompt:

“I’m giving you Google Search Console data showing the search performance of a URL. Can you analyze?”

ChatGPT replies with the five columns of data and suggested areas it could analyze (e.g., queries generating the most clicks).

Since any ranking page also ranks for a bunch of other things, he consolidates the list with this prompt:

“You are an expert SEO who is highly proficient at keyword analysis. Merge the rows with similar queries into single rows with a single, representative marketing keyphrase. As you do so, for each row, combine the data for impressions and clicks and averaging the data for CTR and Avg Position. Provide link to download.”

Any time you ask AI to manipulate data, ask it to provide a link to download the data so you can confirm that it processed the data properly. In this instance, Andy finds the data from ChatGPT looks good.

The chart includes 19 rows of queries related to what should be on a website homepage, accompanied by their total clicks, impressions, CTR, and position.

With the data on what phrases the page ranks for, Andy pastes the content into ChatGPT and makes this prompt:

“Here’s the page for the data you just analyzed. Evaluate this content’s keyword usage and frequency based on the keywords in the dataset.”

ChatGPT replies with a set of steps it will perform, including text preprocessing and keyword extraction:

• Text preprocessing: We’ll clean the webpage to prepare it for analysis.
• Keyphrase extraction: We’ll identify the keyphrases that appear in the content.
• Frequency count: We’ll count how often each keyphrase appears in the content.
• Evaluation: Finally, we’ll assess the usage of keyphrases based on their frequency in the content and their importance in the aggregated Google Search Console data.

Andy refines the prompt to have ChatGPT get more sophisticated in its recommendations (bolded emphasis by Andy):

“Suggest edits to this page that would rank higher by better indicating its relevance for the keyphrases in the dataset. Focus on recommendations for including the phrases that are not used on the page but appear in the dataset. Make suggestions that improve the flow and clarity of the content, as well as keyphrase relevance. Highlight the recommended changes.”

What content edits can improve rankings?

I find the results quite amazing. Some of the recommended changes generated by ChatGPT:

Introduction

Current: Social media streams are crowded, fast-flowing, and very competitive. As social networks pull back on organic reach, showing up is hard, and getting attention is even harder.

Suggested edit: Navigating the crowded and competitive social media video landscape can be daunting. With platforms like Facebook and Instagram curtailing organic reach, the challenge of making your DIY social media videos visible is greater than ever.

Second section: Winning the algorithm

Current: Social media videos are so engaging that the networks are pushing them way to the top of social streams. Facebook, Instagram, and LinkedIn all want us to do more in this all-powerful format. It wins in the algorithms and it wins for attention.

Suggested edit: The power of creating engaging social media videos is so compelling that platforms like Facebook, Instagram, and LinkedIn are giving them preferential treatment. This is why learning how to create captivating social media video content is crucial—it wins both in algorithms and in capturing audience attention.

Third section: Tools and equipment

Current: Your laptop has a camera and mic, but you can do better than that. A little more effort will get you a high-quality video and much better results.

Suggested edit: While your laptop’s built-in camera and mic may suffice for starters, investing in specialized video equipment for social media marketing can significantly boost the quality and engagement of your DIY social media videos.

Which articles convert email subscribers the best?

Over the past 18 years, Andy has written over 575 articles on topics ranging from influencer marketing to web analytics to email marketing. Some articles perform amazingly, while others receive no traffic. Similarly, some articles get people to subscribe to Orbit’s email list, while others generate zero subscribers.

“When you know that difference, you know exactly what to do on social media. You know what to promote and what’s working well,” Andy says.

In the past, he used Excel spreadsheets to analyze the ­­ for every article on his site. These days, he uses ChatGPT.

First, he collects the data from GA4. Under the Engagement section, he selects “Page and screens: Page path + query string” report. To customize it, he clicks on the pencil icon in the upper right and adds these metrics:

• Sessions
• Users
• Conversions

TIP: If you don’t see your metric right away, type its name and select it when it appears.

Since Andy only wanted blogs in his data set, he clicks “Add filter” in the upper left and builds the filter with these attributes:

• Dimension: Page path and screen class
• Match type: contains
• Value: blog

From here, click the export icon in the upper right (i.e., it’s two icons to the left of the pencil icon) and select “Download CSV.”

Next, Andy creates a second CSV file by changing “Value: blog” to “Value: /blog-newsletter-thank-you-page” to collect his email signup thank you page data:

He clicks the “+” button left of the report columns to select a secondary dimension of “Page Referrer.” The resulting report shows the pages users read right before they subscribed to Orbit’s email list. He exports this data as a CSV.

Andy cleans up this file by deleting the nine rows of metadata at the top, so ChatGPT won’t get confused in its analysis.

The two files are fed into ChatGPT to confirm it could analyze them. Next, Andy asks ChatGPT to do the equivalent of an Excel VLOOKUP so users can find things in a table or a range by row. He uses this prompt:

“Map the data from the two files together into one file, as you would do with vlookup, so each row is a URL and the sessions and conversions are shown for each.”

He asks ChatGPT to provide a file to download. ChatGPT generates 17 rows of URLs accompanied by sessions (traffic), users (traffic), conversions, sessions, users (converted), and conversions (newsletter):

Then, Andy prompts ChatGPT to draw a bar chart showing the conversion rate for each blog post.

ChatGPT responds with how it calculates the conversion rate for each blog post and visualizes a bar chart showing the top 20 blog posts by conversion rate.

NOTE: Conversion rate data is only available because ChatGPT divided the conversion data (numerator) by session traffic (denominator).

Next steps

To amplify the articles that drive the most email conversions, Andy suggests:

• Promoting the article(s) in your email signature.
• Featuring them on your homepage.
• Putting them into a social media marketing rotation.
• Linking to them in presentations.
• Linking to them from high-traffic pages.

Which email campaigns perform the best?

You can also use the GA4-ChatGPT combination to understand which email campaigns perform the best. Here’s how Andy does it:

• Export GA4 data showing traffic to pages based on UTM codes related to email campaigns.
• Export data from Orbit’s email marketing platform (ActiveCampaign).
• Ask ChatGPT to merge the two datasets, analyze, and provide recommendations

To collect the GA4 data, Andy uses the traffic acquisition report, which can be found under Acquisition in the left navigation bar. He clicks on the pencil in the upper right to customize it and selects the following metrics before clicking the “apply” button:

• Sessions
• Engagement rate
• Average engagement time per session
• Conversions
• Session conversion rate

Next, he clicks “Add filter,” selects the “Session medium” dimension, then the value of “email.”

Andy changes the primary dimension to “Session campaign.” The resulting report has one row for each email campaign (i.e., traffic statistics based on the utm_campaign parameter). Andy exports this data as a CSV file.

Next, Andy exports a data file from ActiveCampaign, which includes:

• Campaign name
• Number of sends
• Number of opens
• Number of clicks
• Number of unsubscribes
• Number of bounces
• Open rate
• Click rate
• Unsubscribe date
• Bounce rate

Andy uploads each file separately to ChatGPT and confirms it can process each file. He then prompts ChatGPT to merge the datasets:

“These two datasets show the performance of the same campaigns from both the email service provider (active campaign) and the website behavior (google analytics). I want to see the performance of each campaign across both datasets. Merge the Google Analytics data with the ActiveCampaign data, combining ‘session campaign’ with ‘campaign name’ into one column, inferring which campaigns were the same based on names and dates. Provide link to download.”

Once ChatGPT merges the datasets, Andy confirms the data looks good and puts ChatGPT to work.

The basic prompt is:

“You are a skilled digital strategist and email marketing expert. What insights can you gather from the combined dataset?”

Moonshot prompts include:

• “What can you tell me about newsletter campaigns?”
• “What insights can you gather from this data?”
• “Make a chart using an unusual visualization method showing unexpected insights from the data in this dataset.”
• “How would Leonardo da Vinci visualize this data? Show your thinking.”

Next, Andy asks ChatGPT to categorize the newsletter campaigns and provide insights on metrics for each category. It returns these strategic insights:

• Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.
• Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.
• SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.
• Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Report: Acquisition > Traffic acquisition: Search default channel group

Dimension: Session campaign

Secondary dimension: Date

Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).

Reevaluate SEO campaigns.

Enhance personalization.

Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.

Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.

SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.

Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

You’ve used the GA4 version of Google Analytics for at least a year, if not two to three. You’ve also probably tested the potential of ChatGPT and similar generative AI tools.

But have you combined the power of both?

That’s where Andy Crestodina comes in. The co-founder and CMO of Orbit Media Studios discussed the opportunities in his Marketing Analytics & Data Science (MADS) conference presentation, How to Audit Your Content Using GA4: A Data-Driven Approach to SEO and Lead Generation.

In this article, he walks you through step by step how to create GA4 reports and prompt ChatGPT to identify rising and falling search traffic, improve content rankings, and deliver more effective newsletters.

Which content has rising and falling search traffic?

To see traffic from organic search:

• Go to the Engagement section in the left column.
• In the dropdown, click on the report “Pages and screens: Page title and screen class.”
• Next, click “Add filter” to see organic traffic.

To build the filter:

• Go to the Include dropdown in the right column.
• Select Session medium.
• Under Dimension values, select organic.

At the top of the page, click on “Add comparison.” Select the period, such as the last seven days or the last 28 days. Scroll down to turn on the Compare toggle.

Andy says GA4’s date comparison feature is better than the old Universal Analytics because its comparison matches the days of the week between the two periods.

The resulting report identifies rising and falling search traffic as illustrated in this report:

In the report above, the URL — /blog/how-to-write-testimonials-examples — saw a 43.43% drop in views when comparing Jan. 7 through Feb. 3, 2023, to Dec. 10, 2022, through Jan. 6, 2023. But the URL — /blog/website-navigation — saw a 42.86% rise in search traffic during that same period.

Once you’ve identified content with falling search traffic, consider improving it with these ideas:

What phrases does this article rank for?

Knowing the phrases an article ranks for helps you understand why people find the page. It also helps determine how to optimize for additional phrases and drive more organic search traffic to it.

To do this, Andy enlists the help of ChatGPT. However, the first step is to generate a report from Google Search Console:

• Search type: Web
• Date: [a timeframe] (In this example, Andy selected the last 3 months.)
• Query: -orbit (i.e., to exclude branded queries for Andy’s agency Orbit Media)
• Page: [the page URL]

Click “Export” in the upper right to generate a CSV file containing the data.

Feed the CSV file into ChatGPT. Andy uses ChatGPT Plus ($20 a month) because it supports data analysis and file uploads. He starts with this prompt:

“I’m giving you Google Search Console data showing the search performance of a URL. Can you analyze?”

ChatGPT replies with the five columns of data and suggested areas it could analyze (e.g., queries generating the most clicks).

Since any ranking page also ranks for a bunch of other things, he consolidates the list with this prompt:

“You are an expert SEO who is highly proficient at keyword analysis. Merge the rows with similar queries into single rows with a single, representative marketing keyphrase. As you do so, for each row, combine the data for impressions and clicks and averaging the data for CTR and Avg Position. Provide link to download.”

Any time you ask AI to manipulate data, ask it to provide a link to download the data so you can confirm that it processed the data properly. In this instance, Andy finds the data from ChatGPT looks good.

The chart includes 19 rows of queries related to what should be on a website homepage, accompanied by their total clicks, impressions, CTR, and position.

With the data on what phrases the page ranks for, Andy pastes the content into ChatGPT and makes this prompt:

“Here’s the page for the data you just analyzed. Evaluate this content’s keyword usage and frequency based on the keywords in the dataset.”

ChatGPT replies with a set of steps it will perform, including text preprocessing and keyword extraction:

• Text preprocessing: We’ll clean the webpage to prepare it for analysis.
• Keyphrase extraction: We’ll identify the keyphrases that appear in the content.
• Frequency count: We’ll count how often each keyphrase appears in the content.
• Evaluation: Finally, we’ll assess the usage of keyphrases based on their frequency in the content and their importance in the aggregated Google Search Console data.

Andy refines the prompt to have ChatGPT get more sophisticated in its recommendations (bolded emphasis by Andy):

“Suggest edits to this page that would rank higher by better indicating its relevance for the keyphrases in the dataset. Focus on recommendations for including the phrases that are not used on the page but appear in the dataset. Make suggestions that improve the flow and clarity of the content, as well as keyphrase relevance. Highlight the recommended changes.”

What content edits can improve rankings?

I find the results quite amazing. Some of the recommended changes generated by ChatGPT:

Introduction

Current: Social media streams are crowded, fast-flowing, and very competitive. As social networks pull back on organic reach, showing up is hard, and getting attention is even harder.

Suggested edit: Navigating the crowded and competitive social media video landscape can be daunting. With platforms like Facebook and Instagram curtailing organic reach, the challenge of making your DIY social media videos visible is greater than ever.

Second section: Winning the algorithm

Current: Social media videos are so engaging that the networks are pushing them way to the top of social streams. Facebook, Instagram, and LinkedIn all want us to do more in this all-powerful format. It wins in the algorithms and it wins for attention.

Suggested edit: The power of creating engaging social media videos is so compelling that platforms like Facebook, Instagram, and LinkedIn are giving them preferential treatment. This is why learning how to create captivating social media video content is crucial—it wins both in algorithms and in capturing audience attention.

Third section: Tools and equipment

Current: Your laptop has a camera and mic, but you can do better than that. A little more effort will get you a high-quality video and much better results.

Suggested edit: While your laptop’s built-in camera and mic may suffice for starters, investing in specialized video equipment for social media marketing can significantly boost the quality and engagement of your DIY social media videos.

Which articles convert email subscribers the best?

Over the past 18 years, Andy has written over 575 articles on topics ranging from influencer marketing to web analytics to email marketing. Some articles perform amazingly, while others receive no traffic. Similarly, some articles get people to subscribe to Orbit’s email list, while others generate zero subscribers.

“When you know that difference, you know exactly what to do on social media. You know what to promote and what’s working well,” Andy says.

In the past, he used Excel spreadsheets to analyze the ­­ for every article on his site. These days, he uses ChatGPT.

First, he collects the data from GA4. Under the Engagement section, he selects “Page and screens: Page path + query string” report. To customize it, he clicks on the pencil icon in the upper right and adds these metrics:

• Sessions
• Users
• Conversions

TIP: If you don’t see your metric right away, type its name and select it when it appears.

Since Andy only wanted blogs in his data set, he clicks “Add filter” in the upper left and builds the filter with these attributes:

• Dimension: Page path and screen class
• Match type: contains
• Value: blog

From here, click the export icon in the upper right (i.e., it’s two icons to the left of the pencil icon) and select “Download CSV.”

Next, Andy creates a second CSV file by changing “Value: blog” to “Value: /blog-newsletter-thank-you-page” to collect his email signup thank you page data:

He clicks the “+” button left of the report columns to select a secondary dimension of “Page Referrer.” The resulting report shows the pages users read right before they subscribed to Orbit’s email list. He exports this data as a CSV.

Andy cleans up this file by deleting the nine rows of metadata at the top, so ChatGPT won’t get confused in its analysis.

The two files are fed into ChatGPT to confirm it could analyze them. Next, Andy asks ChatGPT to do the equivalent of an Excel VLOOKUP so users can find things in a table or a range by row. He uses this prompt:

“Map the data from the two files together into one file, as you would do with vlookup, so each row is a URL and the sessions and conversions are shown for each.”

He asks ChatGPT to provide a file to download. ChatGPT generates 17 rows of URLs accompanied by sessions (traffic), users (traffic), conversions, sessions, users (converted), and conversions (newsletter):

Then, Andy prompts ChatGPT to draw a bar chart showing the conversion rate for each blog post.

ChatGPT responds with how it calculates the conversion rate for each blog post and visualizes a bar chart showing the top 20 blog posts by conversion rate.

NOTE: Conversion rate data is only available because ChatGPT divided the conversion data (numerator) by session traffic (denominator).

Next steps

To amplify the articles that drive the most email conversions, Andy suggests:

• Promoting the article(s) in your email signature.
• Featuring them on your homepage.
• Putting them into a social media marketing rotation.
• Linking to them in presentations.
• Linking to them from high-traffic pages.

Which email campaigns perform the best?

You can also use the GA4-ChatGPT combination to understand which email campaigns perform the best. Here’s how Andy does it:

• Export GA4 data showing traffic to pages based on UTM codes related to email campaigns.
• Export data from Orbit’s email marketing platform (ActiveCampaign).
• Ask ChatGPT to merge the two datasets, analyze, and provide recommendations

To collect the GA4 data, Andy uses the traffic acquisition report, which can be found under Acquisition in the left navigation bar. He clicks on the pencil in the upper right to customize it and selects the following metrics before clicking the “apply” button:

• Sessions
• Engagement rate
• Average engagement time per session
• Conversions
• Session conversion rate

Next, he clicks “Add filter,” selects the “Session medium” dimension, then the value of “email.”

Andy changes the primary dimension to “Session campaign.” The resulting report has one row for each email campaign (i.e., traffic statistics based on the utm_campaign parameter). Andy exports this data as a CSV file.

Next, Andy exports a data file from ActiveCampaign, which includes:

• Campaign name
• Number of sends
• Number of opens
• Number of clicks
• Number of unsubscribes
• Number of bounces
• Open rate
• Click rate
• Unsubscribe date
• Bounce rate

Andy uploads each file separately to ChatGPT and confirms it can process each file. He then prompts ChatGPT to merge the datasets:

“These two datasets show the performance of the same campaigns from both the email service provider (active campaign) and the website behavior (google analytics). I want to see the performance of each campaign across both datasets. Merge the Google Analytics data with the ActiveCampaign data, combining ‘session campaign’ with ‘campaign name’ into one column, inferring which campaigns were the same based on names and dates. Provide link to download.”

Once ChatGPT merges the datasets, Andy confirms the data looks good and puts ChatGPT to work.

The basic prompt is:

“You are a skilled digital strategist and email marketing expert. What insights can you gather from the combined dataset?”

Moonshot prompts include:

• “What can you tell me about newsletter campaigns?”
• “What insights can you gather from this data?”
• “Make a chart using an unusual visualization method showing unexpected insights from the data in this dataset.”
• “How would Leonardo da Vinci visualize this data? Show your thinking.”

Next, Andy asks ChatGPT to categorize the newsletter campaigns and provide insights on metrics for each category. It returns these strategic insights:

• Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.
• Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.
• SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.
• Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

“Based on this data, what type of campaigns should be sent more often?”

“What types of campaigns should be abandoned?”

“Suggest changes that would improve the efficiency of this email program.”

“Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).

Reevaluate SEO campaigns.

Enhance personalization.

Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.

Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.

SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.

Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

You’ve used the GA4 version of Google Analytics for at least a year, if not two to three. You’ve also probably tested the potential of ChatGPT and similar generative AI tools.

But have you combined the power of both?

That’s where Andy Crestodina comes in. The co-founder and CMO of Orbit Media Studios discussed the opportunities in his Marketing Analytics & Data Science (MADS) conference presentation, How to Audit Your Content Using GA4: A Data-Driven Approach to SEO and Lead Generation.

In this article, he walks you through step by step how to create GA4 reports and prompt ChatGPT to identify rising and falling search traffic, improve content rankings, and deliver more effective newsletters.

Which content has rising and falling search traffic?

To see traffic from organic search:

• Go to the Engagement section in the left column.
• In the dropdown, click on the report “Pages and screens: Page title and screen class.”
• Next, click “Add filter” to see organic traffic.

To build the filter:

• Go to the Include dropdown in the right column.
• Select Session medium.
• Under Dimension values, select organic.

At the top of the page, click on “Add comparison.” Select the period, such as the last seven days or the last 28 days. Scroll down to turn on the Compare toggle.

Andy says GA4’s date comparison feature is better than the old Universal Analytics because its comparison matches the days of the week between the two periods.

The resulting report identifies rising and falling search traffic as illustrated in this report:

In the report above, the URL — /blog/how-to-write-testimonials-examples — saw a 43.43% drop in views when comparing Jan. 7 through Feb. 3, 2023, to Dec. 10, 2022, through Jan. 6, 2023. But the URL — /blog/website-navigation — saw a 42.86% rise in search traffic during that same period.

Once you’ve identified content with falling search traffic, consider improving it with these ideas:

What phrases does this article rank for?

Knowing the phrases an article ranks for helps you understand why people find the page. It also helps determine how to optimize for additional phrases and drive more organic search traffic to it.

To do this, Andy enlists the help of ChatGPT. However, the first step is to generate a report from Google Search Console:

• Search type: Web
• Date: [a timeframe] (In this example, Andy selected the last 3 months.)
• Query: -orbit (i.e., to exclude branded queries for Andy’s agency Orbit Media)
• Page: [the page URL]

Click “Export” in the upper right to generate a CSV file containing the data.

Feed the CSV file into ChatGPT. Andy uses ChatGPT Plus ($20 a month) because it supports data analysis and file uploads. He starts with this prompt:

“I’m giving you Google Search Console data showing the search performance of a URL. Can you analyze?”

ChatGPT replies with the five columns of data and suggested areas it could analyze (e.g., queries generating the most clicks).

Since any ranking page also ranks for a bunch of other things, he consolidates the list with this prompt:

“You are an expert SEO who is highly proficient at keyword analysis. Merge the rows with similar queries into single rows with a single, representative marketing keyphrase. As you do so, for each row, combine the data for impressions and clicks and averaging the data for CTR and Avg Position. Provide link to download.”

Any time you ask AI to manipulate data, ask it to provide a link to download the data so you can confirm that it processed the data properly. In this instance, Andy finds the data from ChatGPT looks good.

The chart includes 19 rows of queries related to what should be on a website homepage, accompanied by their total clicks, impressions, CTR, and position.

With the data on what phrases the page ranks for, Andy pastes the content into ChatGPT and makes this prompt:

“Here’s the page for the data you just analyzed. Evaluate this content’s keyword usage and frequency based on the keywords in the dataset.”

ChatGPT replies with a set of steps it will perform, including text preprocessing and keyword extraction:

• Text preprocessing: We’ll clean the webpage to prepare it for analysis.
• Keyphrase extraction: We’ll identify the keyphrases that appear in the content.
• Frequency count: We’ll count how often each keyphrase appears in the content.
• Evaluation: Finally, we’ll assess the usage of keyphrases based on their frequency in the content and their importance in the aggregated Google Search Console data.

Andy refines the prompt to have ChatGPT get more sophisticated in its recommendations (bolded emphasis by Andy):

“Suggest edits to this page that would rank higher by better indicating its relevance for the keyphrases in the dataset. Focus on recommendations for including the phrases that are not used on the page but appear in the dataset. Make suggestions that improve the flow and clarity of the content, as well as keyphrase relevance. Highlight the recommended changes.”

What content edits can improve rankings?

I find the results quite amazing. Some of the recommended changes generated by ChatGPT:

Introduction

Current: Social media streams are crowded, fast-flowing, and very competitive. As social networks pull back on organic reach, showing up is hard, and getting attention is even harder.

Suggested edit: Navigating the crowded and competitive social media video landscape can be daunting. With platforms like Facebook and Instagram curtailing organic reach, the challenge of making your DIY social media videos visible is greater than ever.

Second section: Winning the algorithm

Current: Social media videos are so engaging that the networks are pushing them way to the top of social streams. Facebook, Instagram, and LinkedIn all want us to do more in this all-powerful format. It wins in the algorithms and it wins for attention.

Suggested edit: The power of creating engaging social media videos is so compelling that platforms like Facebook, Instagram, and LinkedIn are giving them preferential treatment. This is why learning how to create captivating social media video content is crucial—it wins both in algorithms and in capturing audience attention.

Third section: Tools and equipment

Current: Your laptop has a camera and mic, but you can do better than that. A little more effort will get you a high-quality video and much better results.

Suggested edit: While your laptop’s built-in camera and mic may suffice for starters, investing in specialized video equipment for social media marketing can significantly boost the quality and engagement of your DIY social media videos.

Which articles convert email subscribers the best?

Over the past 18 years, Andy has written over 575 articles on topics ranging from influencer marketing to web analytics to email marketing. Some articles perform amazingly, while others receive no traffic. Similarly, some articles get people to subscribe to Orbit’s email list, while others generate zero subscribers.

“When you know that difference, you know exactly what to do on social media. You know what to promote and what’s working well,” Andy says.

In the past, he used Excel spreadsheets to analyze the ­­ for every article on his site. These days, he uses ChatGPT.

First, he collects the data from GA4. Under the Engagement section, he selects “Page and screens: Page path + query string” report. To customize it, he clicks on the pencil icon in the upper right and adds these metrics:

• Sessions
• Users
• Conversions

TIP: If you don’t see your metric right away, type its name and select it when it appears.

Since Andy only wanted blogs in his data set, he clicks “Add filter” in the upper left and builds the filter with these attributes:

• Dimension: Page path and screen class
• Match type: contains
• Value: blog

From here, click the export icon in the upper right (i.e., it’s two icons to the left of the pencil icon) and select “Download CSV.”

Next, Andy creates a second CSV file by changing “Value: blog” to “Value: /blog-newsletter-thank-you-page” to collect his email signup thank you page data:

He clicks the “+” button left of the report columns to select a secondary dimension of “Page Referrer.” The resulting report shows the pages users read right before they subscribed to Orbit’s email list. He exports this data as a CSV.

Andy cleans up this file by deleting the nine rows of metadata at the top, so ChatGPT won’t get confused in its analysis.

The two files are fed into ChatGPT to confirm it could analyze them. Next, Andy asks ChatGPT to do the equivalent of an Excel VLOOKUP so users can find things in a table or a range by row. He uses this prompt:

“Map the data from the two files together into one file, as you would do with vlookup, so each row is a URL and the sessions and conversions are shown for each.”

He asks ChatGPT to provide a file to download. ChatGPT generates 17 rows of URLs accompanied by sessions (traffic), users (traffic), conversions, sessions, users (converted), and conversions (newsletter):

Then, Andy prompts ChatGPT to draw a bar chart showing the conversion rate for each blog post.

ChatGPT responds with how it calculates the conversion rate for each blog post and visualizes a bar chart showing the top 20 blog posts by conversion rate.

NOTE: Conversion rate data is only available because ChatGPT divided the conversion data (numerator) by session traffic (denominator).

Next steps

To amplify the articles that drive the most email conversions, Andy suggests:

• Promoting the article(s) in your email signature.
• Featuring them on your homepage.
• Putting them into a social media marketing rotation.
• Linking to them in presentations.
• Linking to them from high-traffic pages.

Which email campaigns perform the best?

You can also use the GA4-ChatGPT combination to understand which email campaigns perform the best. Here’s how Andy does it:

• Export GA4 data showing traffic to pages based on UTM codes related to email campaigns.
• Export data from Orbit’s email marketing platform (ActiveCampaign).
• Ask ChatGPT to merge the two datasets, analyze, and provide recommendations

To collect the GA4 data, Andy uses the traffic acquisition report, which can be found under Acquisition in the left navigation bar. He clicks on the pencil in the upper right to customize it and selects the following metrics before clicking the “apply” button:

• Sessions
• Engagement rate
• Average engagement time per session
• Conversions
• Session conversion rate

Next, he clicks “Add filter,” selects the “Session medium” dimension, then the value of “email.”

Andy changes the primary dimension to “Session campaign.” The resulting report has one row for each email campaign (i.e., traffic statistics based on the utm_campaign parameter). Andy exports this data as a CSV file.

Next, Andy exports a data file from ActiveCampaign, which includes:

• Campaign name
• Number of sends
• Number of opens
• Number of clicks
• Number of unsubscribes
• Number of bounces
• Open rate
• Click rate
• Unsubscribe date
• Bounce rate

Andy uploads each file separately to ChatGPT and confirms it can process each file. He then prompts ChatGPT to merge the datasets:

“These two datasets show the performance of the same campaigns from both the email service provider (active campaign) and the website behavior (google analytics). I want to see the performance of each campaign across both datasets. Merge the Google Analytics data with the ActiveCampaign data, combining ‘session campaign’ with ‘campaign name’ into one column, inferring which campaigns were the same based on names and dates. Provide link to download.”

Once ChatGPT merges the datasets, Andy confirms the data looks good and puts ChatGPT to work.

The basic prompt is:

“You are a skilled digital strategist and email marketing expert. What insights can you gather from the combined dataset?”

Moonshot prompts include:

• “What can you tell me about newsletter campaigns?”
• “What insights can you gather from this data?”
• “Make a chart using an unusual visualization method showing unexpected insights from the data in this dataset.”
• “How would Leonardo da Vinci visualize this data? Show your thinking.”

Next, Andy asks ChatGPT to categorize the newsletter campaigns and provide insights on metrics for each category. It returns these strategic insights:

• Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.
• Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.
• SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.
• Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Report: Acquisition > Traffic acquisition: Search default channel group

Dimension: Session campaign

Secondary dimension: Date

Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

“Based on this data, what type of campaigns should be sent more often?”

“What types of campaigns should be abandoned?”

“Suggest changes that would improve the efficiency of this email program.”

“Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).

Reevaluate SEO campaigns.

Enhance personalization.

Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.

Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.

SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.

Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute

You’ve used the GA4 version of Google Analytics for at least a year, if not two to three. You’ve also probably tested the potential of ChatGPT and similar generative AI tools.

But have you combined the power of both?

That’s where Andy Crestodina comes in. The co-founder and CMO of Orbit Media Studios discussed the opportunities in his Marketing Analytics & Data Science (MADS) conference presentation, How to Audit Your Content Using GA4: A Data-Driven Approach to SEO and Lead Generation.

In this article, he walks you through step by step how to create GA4 reports and prompt ChatGPT to identify rising and falling search traffic, improve content rankings, and deliver more effective newsletters.

Which content has rising and falling search traffic?

To see traffic from organic search:

• Go to the Engagement section in the left column.
• In the dropdown, click on the report “Pages and screens: Page title and screen class.”
• Next, click “Add filter” to see organic traffic.

To build the filter:

• Go to the Include dropdown in the right column.
• Select Session medium.
• Under Dimension values, select organic.

At the top of the page, click on “Add comparison.” Select the period, such as the last seven days or the last 28 days. Scroll down to turn on the Compare toggle.

Andy says GA4’s date comparison feature is better than the old Universal Analytics because its comparison matches the days of the week between the two periods.

The resulting report identifies rising and falling search traffic as illustrated in this report:

In the report above, the URL — /blog/how-to-write-testimonials-examples — saw a 43.43% drop in views when comparing Jan. 7 through Feb. 3, 2023, to Dec. 10, 2022, through Jan. 6, 2023. But the URL — /blog/website-navigation — saw a 42.86% rise in search traffic during that same period.

Once you’ve identified content with falling search traffic, consider improving it with these ideas:

What phrases does this article rank for?

Knowing the phrases an article ranks for helps you understand why people find the page. It also helps determine how to optimize for additional phrases and drive more organic search traffic to it.

To do this, Andy enlists the help of ChatGPT. However, the first step is to generate a report from Google Search Console:

• Search type: Web
• Date: [a timeframe] (In this example, Andy selected the last 3 months.)
• Query: -orbit (i.e., to exclude branded queries for Andy’s agency Orbit Media)
• Page: [the page URL]

Click “Export” in the upper right to generate a CSV file containing the data.

Feed the CSV file into ChatGPT. Andy uses ChatGPT Plus ($20 a month) because it supports data analysis and file uploads. He starts with this prompt:

“I’m giving you Google Search Console data showing the search performance of a URL. Can you analyze?”

ChatGPT replies with the five columns of data and suggested areas it could analyze (e.g., queries generating the most clicks).

Since any ranking page also ranks for a bunch of other things, he consolidates the list with this prompt:

“You are an expert SEO who is highly proficient at keyword analysis. Merge the rows with similar queries into single rows with a single, representative marketing keyphrase. As you do so, for each row, combine the data for impressions and clicks and averaging the data for CTR and Avg Position. Provide link to download.”

Any time you ask AI to manipulate data, ask it to provide a link to download the data so you can confirm that it processed the data properly. In this instance, Andy finds the data from ChatGPT looks good.

The chart includes 19 rows of queries related to what should be on a website homepage, accompanied by their total clicks, impressions, CTR, and position.

With the data on what phrases the page ranks for, Andy pastes the content into ChatGPT and makes this prompt:

“Here’s the page for the data you just analyzed. Evaluate this content’s keyword usage and frequency based on the keywords in the dataset.”

ChatGPT replies with a set of steps it will perform, including text preprocessing and keyword extraction:

• Text preprocessing: We’ll clean the webpage to prepare it for analysis.
• Keyphrase extraction: We’ll identify the keyphrases that appear in the content.
• Frequency count: We’ll count how often each keyphrase appears in the content.
• Evaluation: Finally, we’ll assess the usage of keyphrases based on their frequency in the content and their importance in the aggregated Google Search Console data.

Andy refines the prompt to have ChatGPT get more sophisticated in its recommendations (bolded emphasis by Andy):

“Suggest edits to this page that would rank higher by better indicating its relevance for the keyphrases in the dataset. Focus on recommendations for including the phrases that are not used on the page but appear in the dataset. Make suggestions that improve the flow and clarity of the content, as well as keyphrase relevance. Highlight the recommended changes.”

What content edits can improve rankings?

I find the results quite amazing. Some of the recommended changes generated by ChatGPT:

Introduction

Current: Social media streams are crowded, fast-flowing, and very competitive. As social networks pull back on organic reach, showing up is hard, and getting attention is even harder.

Suggested edit: Navigating the crowded and competitive social media video landscape can be daunting. With platforms like Facebook and Instagram curtailing organic reach, the challenge of making your DIY social media videos visible is greater than ever.

Second section: Winning the algorithm

Current: Social media videos are so engaging that the networks are pushing them way to the top of social streams. Facebook, Instagram, and LinkedIn all want us to do more in this all-powerful format. It wins in the algorithms and it wins for attention.

Suggested edit: The power of creating engaging social media videos is so compelling that platforms like Facebook, Instagram, and LinkedIn are giving them preferential treatment. This is why learning how to create captivating social media video content is crucial—it wins both in algorithms and in capturing audience attention.

Third section: Tools and equipment

Current: Your laptop has a camera and mic, but you can do better than that. A little more effort will get you a high-quality video and much better results.

Suggested edit: While your laptop’s built-in camera and mic may suffice for starters, investing in specialized video equipment for social media marketing can significantly boost the quality and engagement of your DIY social media videos.

Which articles convert email subscribers the best?

Over the past 18 years, Andy has written over 575 articles on topics ranging from influencer marketing to web analytics to email marketing. Some articles perform amazingly, while others receive no traffic. Similarly, some articles get people to subscribe to Orbit’s email list, while others generate zero subscribers.

“When you know that difference, you know exactly what to do on social media. You know what to promote and what’s working well,” Andy says.

In the past, he used Excel spreadsheets to analyze the ­­ for every article on his site. These days, he uses ChatGPT.

First, he collects the data from GA4. Under the Engagement section, he selects “Page and screens: Page path + query string” report. To customize it, he clicks on the pencil icon in the upper right and adds these metrics:

• Sessions
• Users
• Conversions

TIP: If you don’t see your metric right away, type its name and select it when it appears.

Since Andy only wanted blogs in his data set, he clicks “Add filter” in the upper left and builds the filter with these attributes:

• Dimension: Page path and screen class
• Match type: contains
• Value: blog

From here, click the export icon in the upper right (i.e., it’s two icons to the left of the pencil icon) and select “Download CSV.”

Next, Andy creates a second CSV file by changing “Value: blog” to “Value: /blog-newsletter-thank-you-page” to collect his email signup thank you page data:

He clicks the “+” button left of the report columns to select a secondary dimension of “Page Referrer.” The resulting report shows the pages users read right before they subscribed to Orbit’s email list. He exports this data as a CSV.

Andy cleans up this file by deleting the nine rows of metadata at the top, so ChatGPT won’t get confused in its analysis.

The two files are fed into ChatGPT to confirm it could analyze them. Next, Andy asks ChatGPT to do the equivalent of an Excel VLOOKUP so users can find things in a table or a range by row. He uses this prompt:

“Map the data from the two files together into one file, as you would do with vlookup, so each row is a URL and the sessions and conversions are shown for each.”

He asks ChatGPT to provide a file to download. ChatGPT generates 17 rows of URLs accompanied by sessions (traffic), users (traffic), conversions, sessions, users (converted), and conversions (newsletter):

Then, Andy prompts ChatGPT to draw a bar chart showing the conversion rate for each blog post.

ChatGPT responds with how it calculates the conversion rate for each blog post and visualizes a bar chart showing the top 20 blog posts by conversion rate.

NOTE: Conversion rate data is only available because ChatGPT divided the conversion data (numerator) by session traffic (denominator).

Next steps

To amplify the articles that drive the most email conversions, Andy suggests:

• Promoting the article(s) in your email signature.
• Featuring them on your homepage.
• Putting them into a social media marketing rotation.
• Linking to them in presentations.
• Linking to them from high-traffic pages.

Which email campaigns perform the best?

You can also use the GA4-ChatGPT combination to understand which email campaigns perform the best. Here’s how Andy does it:

• Export GA4 data showing traffic to pages based on UTM codes related to email campaigns.
• Export data from Orbit’s email marketing platform (ActiveCampaign).
• Ask ChatGPT to merge the two datasets, analyze, and provide recommendations

To collect the GA4 data, Andy uses the traffic acquisition report, which can be found under Acquisition in the left navigation bar. He clicks on the pencil in the upper right to customize it and selects the following metrics before clicking the “apply” button:

• Sessions
• Engagement rate
• Average engagement time per session
• Conversions
• Session conversion rate

Next, he clicks “Add filter,” selects the “Session medium” dimension, then the value of “email.”

Andy changes the primary dimension to “Session campaign.” The resulting report has one row for each email campaign (i.e., traffic statistics based on the utm_campaign parameter). Andy exports this data as a CSV file.

Next, Andy exports a data file from ActiveCampaign, which includes:

• Campaign name
• Number of sends
• Number of opens
• Number of clicks
• Number of unsubscribes
• Number of bounces
• Open rate
• Click rate
• Unsubscribe date
• Bounce rate

Andy uploads each file separately to ChatGPT and confirms it can process each file. He then prompts ChatGPT to merge the datasets:

“These two datasets show the performance of the same campaigns from both the email service provider (active campaign) and the website behavior (google analytics). I want to see the performance of each campaign across both datasets. Merge the Google Analytics data with the ActiveCampaign data, combining ‘session campaign’ with ‘campaign name’ into one column, inferring which campaigns were the same based on names and dates. Provide link to download.”

Once ChatGPT merges the datasets, Andy confirms the data looks good and puts ChatGPT to work.

The basic prompt is:

“You are a skilled digital strategist and email marketing expert. What insights can you gather from the combined dataset?”

Moonshot prompts include:

• “What can you tell me about newsletter campaigns?”
• “What insights can you gather from this data?”
• “Make a chart using an unusual visualization method showing unexpected insights from the data in this dataset.”
• “How would Leonardo da Vinci visualize this data? Show your thinking.”

Next, Andy asks ChatGPT to categorize the newsletter campaigns and provide insights on metrics for each category. It returns these strategic insights:

• Engagement with analytics: Campaigns related to “Analytics” have the highest engagement rate (49.79%) and the longest average engagement time (81.51 seconds). This suggests subscribers are deeply interested in analytics topics and spend more time engaging with such content.
• Content marketing effectiveness: While “Content Marketing” has one of the higher open rates (20%), the engagement time is relatively short (35.48 seconds). This might indicate that while the topic attracts initial interest, the content might not be keeping users engaged for long.
• SEO’s lower engagement: Campaigns related to “SEO” have one of the lowest engagement rates and shortest engagement times. It might be worth revisiting the content or presentation style for these campaigns to boost engagement.
• Email marketing’s extended engagement: Despite having an average open rate, “Email Marketing” topics keep users engaged for a longer duration (63.94 seconds), indicating the content’s depth or relevance to the audience.

Andy asks ChatGPT to suggest adjustments to his email marketing program based on the data provided. High-level suggestions include:

• Focus on high-engagement topics (e.g., analytics, content marketing, email marketing).
• Reevaluate SEO campaigns.
• Enhance personalization.
• Optimize send times.

Each high-level suggestion includes supporting details and leads to additional prompts from Andy, including:

• “Based on this data, what type of campaigns should be sent more often?”
• “What types of campaigns should be abandoned?”
• “Suggest changes that would improve the efficiency of this email program.”
• “Based on this data, what five possible subject lines would have the highest website conversion rates?”

Campaign timing analysis (dates, days, and seasonality)

In his 20-plus years doing web analytics, Andy has never generated a report using the newsletter’s date as a secondary dimension. However, generative AI makes it possible to find patterns in timing-related data.

To start, Andy generates a GA4 report showing traffic acquisition from his newsletter, sorted by date:

• Report: Acquisition > Traffic acquisition: Search default channel group
• Dimension: Session campaign
• Secondary dimension: Date
• Filter: Source medium contains “newsletter”

Andy gives the report to ChatGPT and makes this prompt:

“Draw two charts. One showing correlations by month. One showing correlations by day-of-week. Normalize the data.”

The month correlations chart shows strong performance in the early months of the year, a drop-through October, and a rise in November and December to the early-year level. In the day-of-week chart, Thursday is the clear winner, and Saturday is the clear loser.

The next prompt calls for sophisticated analysis:

“Create and display a one-year calendar for this newsletter. Schedule it bi-weekly, selecting dates for optimal performance. Write draft headlines for each, selecting topics for optimal performance.”

Among ChatGPT’s suggestions:

• Optimal day: Friday
• Optimal topics: Content marketing, SEO
• Start date: Upcoming Friday
• Frequency: Every other Friday

Of course, you don’t need to follow ChatGPT’s suggestions exactly. However, I’m fascinated that it can generate these suggestions based on the data Andy provided.

Complementing and augmenting analytics and AI

I was competent in Google’s Universal Analytics, but I’m a complete newbie in how to marry GA4 reports with generative AI tools. Andy provides a nice pathway for non-experts:

• Create your go-to reports in GA4.
• Export the data into CSV files.
• Feed the files into a generative AI tool like ChatGPT.
• Prompt the tool to analyze the data and provide recommendations.

With that process, you can learn how to optimize existing content to improve rankings, convert more visitors, and create content calendars that detail the most effective headlines, topics, frequency, and distribution days.

I now turn things over to you and ask: What will you ask generative AI to do for your marketing initiatives?

Let me know on social media using the tag #CMWorld.

All tools mentioned in this article were suggested by the author. If you’d like to suggest a tool, share the article on social media with a comment.

HANDPICKED RELATED CONTENT:

Cover image by Joseph Kalinowski/Content Marketing Institute