I used GA4 data and ChatGPT-4 to measurably improve my website content.
I uploaded my page-level GA4 data into ChatGPT-4, and used deep research to find gaps and unanswered questions in my content about GA4 training for government teams.
It blew my mind.
What impressed me was the way the report combined analysis of my actual content with marketing and audience insights. It provided very tailored recommendations, using legit GA4 insights (see examples below).
I followed the recommendations to create a new page, and the results are good.
Average engagement time:
New page: 63 seconds
Other Service pages: 24 seconds
BUT, there are considerations:
- There were too many recommendations and too much detail, ChatGPT-4 is an over-enthusiastic option-gatherer without the experience to know what actually works.
- I didn’t trust AI to write the content for me, it took a lot of my experience to work out what suggestions to apply and how.
- Chat GPT4 didn’t factor in GA4 data accuracy issues that could affect performance insights.
- This is a basic site, and I made a basic content improvement request. I don’t think AI is ready to handle the complex performance needs yet.
And please check whether the owner of your GA4 data is happy for you to upload it into GA4 – check the organisation’s AI policy.
ChatGPT-4 strengths
AI is transforming marketing. I see it super-charging in-house marketing for small-medium sized businesses who can start small and test things, and keep an eye on the overall content and IA strategy.
It’s exciting that large organisations with complex sites will be able to use AI to meet user needs better – it’s great at finding user insight trends in data sets, and from research across the internet.
ChatGPT-4 weaknesses
It has a while to go before it can analyse sites with complex performance. There’s also a risk that AI will amplify inconsistent use of data, and encourage piecemeal changes to IA and content.
What I did
I uploaded my GA4 Pages and screens report into Chat GPT4, and asked it to review my GA4 page performance data, and give content recommendations on how to market my GA4 training to government agencies.
I used deep research which took 15 minutes and generated a long report, with lots of very tailored recommendations on how to improve my existing content based on GA4 insights, marketing 101 tactics and the government persona insights it had found across the internet.
A key recommendation was to create a lead generation page, with suggestions on what to include in this page.
Prompt – find content gaps and give suggestions to fill them
I wanted to find content gaps, and get ideas on how to fill these. Here’s my prompt, (based on Andy Crestodina’s guidance):
“This is a content performance report from ga4. You are an expert in content optimisation. My goal is to generate leads for my GA4 training and support. Based on a government client persona and the GA4 data provided, identify the most important unanswered questions and key objections or concerns this content fails to address. Provide actionable suggestions to resolve these gaps.”
Report – WOW but too much information
The report took 15 minutes to create, and my brain often short-circuited trying to assess it. I used my experience in analytics, UX, content and knowledge of client needs, to work out which recommendations to apply. I created a new page ‘GA4 training for government teams’ and wrote the content myself, which is now on the GA4 training for teams page, pretty much unchanged.
GA4 + content analysis
The GA4 analysis was good, factoring in the limited data it had to use and the simplicity of my site. It worked mainly off page views and average engagement time on page:
Here’s an excerpt from the ChatGPT-4 content analysis:
“GA4 Pros and Cons – Although it had moderate traffic (48 views), it achieved one of the longest engagement times per user (≈41 seconds on average). This implies readers value in-depth, candid discussion of GA4’s advantages and drawbacks.”
Client (persona) needs
I asked what government teams need from GA4, and initially I was impressed. It picked up things that I hadn’t addressed in my content, like concerns around data privacy, procurement needs, and GA4 vs UA transition issues.
I was surprised to see only two reference sources for this insight however, and would want to test content with my target audience if this was high stakes content.
Tailored content suggestions
There was good stuff there, with 101-level marketing strategies applied in lots of very tailored recommendations. For example:
“For example, bring in a short government case study snippet on the GA4 services page. You might highlight the Archives New Zealand case since it drew interest”.
There were dodgy bits. It suggested content that had no factual basis. Just as an example, but still… This is what it suggested:
“…after working with us, [Organisation name] was able to confidently measure content engagement… helping them secure ongoing support for data-driven projects.”
There were too many recommendations to reasonably apply. The report had a helpful summary of the top priorities, but even these required more work than I was prepared to do.
Result
After just over a month the page has had a few visits, mostly from social promotion. What impresses me is the high engagement time. Most of my service pages are glanced at, but people stay on this one for around a minute.
A few things make me question the results:
- Internal traffic
- SPAM
- If it’s the last page visited, then no engagement time is recorded.
- The page has only been seen by 20 people.
So I’m not shouting too loudly just yet. But I think it’s a strong start, and I’ll keep using AI to inform my website content.
What does ChatGPT-4 think about my changes?
When I uploaded the post-change GA4 data into ChatGPT-4, it was generally positive. I listed the pages I’d added or changed and asked for a review of performance:
Then it gave me an overall summary of the changes. Positive and succinct, if a little premature:
Considerations
Chat GPT4 didn’t pick up on potential data accuracy issues in the GA4 reports.
The GA4 data wasn’t totally reliable due to spam, my own usage data, quirks around how engagement time data is captured, and low traffic levels. I’m not a data purist because GA4 data will never be totally accurate. But I’d want more data before I claimed success.
I didn’t trust AI to write the content for me.
I feel like I can tell when content has been written by AI, when after reading for a while my eyes glaze over. I don’t think AI has learnt what solutions will actually work from all the flotsam and jetsam of internet guidance yet, and it shows. While I use AI to create my titles and service descriptions, I don’t use it to write deep-level content.
I had to use a lot of my experience with GA4 data, UX, content design and my knowledge of my government clients to work out which recommendations to include, and which to discard.
This is a basic site, with a basic content improvement request.
In this example, simple page engagement measures (page views and page engagement time) were enough to inform changes. For large, content driven sites the performance side of things gets complex, with success measured by things like download files, clicks on navigation links, and video engagement.
I don’t think AI is capable of navigating complex performance needs just yet.
Verdict
I was really impressed with ChatGPT-4’s capability when it came to applying GA4 insights to content improvements. The report was much more nuanced and tailored than I thought it would be, and it will transform marketing activities, making it easier to create user-focussed content and improve performance.
ChatGPT-4 is great for finding themes in user data and research, it will scour your data and the internet to find nuanced insights and suggest recommendations.
In a large organisation there’s the potential for inconsistent and piecemeal changes if everyone uses AI to work on isolated parts of the website.
And it’s not so good at understanding the performance of complex websites, yet.