How I Analyse My Site's AI SEO Performance Using Free Tools: ChatGPT + GSC (Generative AI Features) + GA4

Traditional rank tracking still has its place, but it is no longer enough on its own. It can tell you where a page appears in the standard Google results, but it will not accurately tell you whether your content is being cited in Google AI Overviews, ChatGPT, Perplexity or other AI-led search tools.

For website owners and marketers, this can be a critical blind spot. Studies now suggest that AI-generated search results can reduce clicks to normal organic listings, while some AI referral traffic may arrive with stronger intent. The numbers vary by source and sector, so I would be careful about treating any single statistic as universal, but the direction is clear enough, meaning rankings, clicks and visibility are starting to separate.

In this guide, I’ll show how I review AI visibility using Google Search Console, GA4 and ChatGPT. It will not replace a dedicated AI visibility platform, but it gives you a practical starting point using data you probably already have.

Why Your Rank Tracker Is Missing Half the Picture

Most SEO tools were built around a fairly simple model. They check a search result, find your URL, and report the position. For many years, that was a perfectly sensible way to monitor SEO performance.

The problem is that the search results are no longer just a static list of links. Google AI Overviews, featured snippets, People Also Ask, Reddit discussions, shopping modules, videos and paid results can all sit around or above the standard organic results. On top of that, users are increasingly asking tools like ChatGPT and Perplexity questions directly, rather than searching Google in the traditional way.

This matters because AI-generated answers change the relationship between ranking and traffic. Ahrefs found that the presence of an AI Overview correlated with a 58% lower average click-through rate for the top-ranking page in the queries they analysed [1]. Pew Research Center also found that users who saw an AI summary in Google were less likely to click on traditional search result links than users who did not see one [2]. This means we are no longer just looking at ranking, but we are also looking at whether a page is being used as a trusted source.

A traditional rank tracker can tell you that a page ranks third. It cannot tell you whether an AI system has decided to use that page as evidence in its answer, whether a competitor has been cited instead, or whether the answer is satisfying the user before they ever reach your site.

Ranking is about visibility in a list, whereas citation is about being trusted enough to support the generated answer. The two overlap, especially where E-E-A-T is concerned, but they are not the same.

Gartner previously predicted that traditional search engine volume would fall by 25% by 2026 [3]. This is not replacing conventional SEO overnight, but it is already large enough to be worth measuring.

For most businesses, the sensible approach is to add another layer of analysis alongside rank tracking. You still need to know where you rank, but you also need to understand whether your pages are being cited, whether AI results are suppressing clicks, and which pages are already attracting AI referral traffic.

Step 1: Export Your GSC Data and Run It Through ChatGPT

Google Search Console is still one of the most useful tools for this kind of analysis. Most people use it to check clicks, impressions and average positions, but the real value is in the raw query data.

When you export the Queries report, you get a list of the searches your site appeared for, along with impressions, clicks, click-through rate and average position. ChatGPT can help by grouping similar searches, identifying patterns and highlighting areas that deserve closer attention.

How to get the data

1. Log in to Google Search Console
2. Navigate to Performance then Search Results
3. Set the date range to the last three to six months
4. Click Export and choose Download CSV (make sure you have the Queries tab selected)
5. Open the CSV in a plain text editor like Notepad, TextEdit or VS Code. This helps preserve the raw format when pasting into ChatGPT.
6. Copy the contents and paste into ChatGPT with one of the prompts below or attach as a file and reference in the prompt.

Note: GSC exports can be large. If ChatGPT hits its context limit, trim the CSV to the top 200 to 300 rows by impressions before pasting. Larger sites may need to split the export into sections, for example by query type, page, country or date range.

Prompt 1: Identifying AI Overview triggers

Question-based searches are often the ones most likely to trigger AI Overviews, featured snippets or other answer-led search features. This prompt helps identify those queries within your own GSC data.

Analyse this Google Search Console query data. Identify which queries are likely
triggering AI Overviews or featured snippets based on their phrasing (question-based,
definition-based, or comparison-based queries). Group them and list the top 10 most
at-risk queries where I may be losing clicks to AI-generated answers.
[paste CSV data here]

The aim here is not to treat the query export as proof that an AI Overview appeared for every search. That is where Google’s new Generative AI features report now changes the process.

If you have access to it, go to:

Performance > Generative AI

This report shows impressions from Google’s generative AI features, including AI Overviews and AI Mode. In the current version, you can review the data by pages, countries, devices and dates. That makes it far more useful than guessing from normal GSC data alone, because you can see which URLs are actually appearing in Google’s generative AI results.

There are still limits though. At the time of writing, the report does not give the same query-level, click, CTR and average position detail that you get in the standard Search Results report. It is also still being rolled out, so not every site will see it yet [7] [8].

For that reason, I would now use the two reports together:

1. Use Generative AI features to find the URLs already appearing in AI Overviews and AI Mode.
2. Use the standard Search Results > Queries export to understand the likely searches and intents behind those URLs.
3. Use ChatGPT to group those queries into topics, risks and content opportunities.

Once you have that list, check the Generative AI features report to see whether the relevant page is already gaining impressions. If it is, you know the page has some level of visibility in Google’s AI features. If it is not, you still have a useful shortlist of queries to check manually and pages to improve.

Optional Prompt: Cross-checking GSC query data against Generative AI pages

If you export the pages from the Generative AI features report and also export the standard GSC query data, you can ask ChatGPT to compare the two.

I have two Google Search Console exports.
Export 1 contains pages receiving impressions in the Generative AI features report.
Export 2 contains standard Search Results query data with queries, pages, clicks, impressions, CTR and average position.
Compare the two exports and identify:
1. Which pages appear in Generative AI features and which standard search queries are most likely connected to those pages.
2. Which topics or query intents appear to be driving the Generative AI impressions.
3. Which high-impression pages are not appearing in the Generative AI report but may be good candidates for improvement.
4. Any pages where traditional rankings look strong but AI visibility appears weak.
[paste both exports here]

Prompt 2: Finding suppressed clicks

High impressions, reasonable rankings and poor click-through rates can still be a useful warning sign. It does not always mean an AI Overview is responsible. Paid results, map packs, featured snippets, Reddit results, shopping modules and brand bias can all affect CTR as well.

From this Google Search Console data, identify queries where my average position
is between 4 and 15 but impressions are high and CTR is unusually low (below 2%).
These may be queries where an AI Overview is appearing above my result and
suppressing clicks. List them ranked by impression volume.
[paste CSV data here]

I would treat this prompt as a diagnostic check rather than a final answer. It helps separate rankings from outcomes. A page can hold a reasonable average position and still fail to attract the traffic you would normally expect.

The difference now is that you can cross-check any suspicious pages against the Generative AI features report:

• If the page has strong rankings, weak CTR and high Generative AI impressions, it may be appearing in AI results but not earning the click.
• If the page has strong rankings, weak CTR and no Generative AI impressions, another SERP feature may be suppressing clicks.
• If the page has no Generative AI impressions but covers an informational topic, it may need clearer structure, better supporting evidence or stronger topical depth.

This is more useful than simply saying “AI Overviews are stealing clicks”. Sometimes they will be part of the issue. Sometimes they will not. The point is to use the data to decide what deserves closer attention.

Prompt 3: Discovering semantic clusters

Semantic clusters show where multiple searches are really asking the same underlying question. This is useful for AI search because AI systems are less interested in exact-match keyword repetition and more interested in whether a page covers the topic clearly and comprehensively.

Analyse this query data and identify semantic clusters. These are groups of queries that
share the same underlying intent. For each cluster, suggest one piece of content
that would address the full cluster rather than individual keywords.
[paste CSV data here]

For example, searches around “how to optimise images for web”, “image compression”, “WebP images”, “lazy loading images” and “image file size SEO” may all belong to the same broader topic. A thin article targeting one phrase is unlikely to perform as well as a well-structured guide that answers the full set of related questions.

This is where GSC and ChatGPT work well together. GSC tells you what people are actually searching for. The Generative AI features report shows which pages are appearing in Google’s AI-led results. ChatGPT helps group the searches into themes, so you can decide whether to improve an existing page, merge weaker content, or create something new.

Step 2: Use GA4 to Find Your AI Referral Traffic

GSC now gives a direct view of Google generative AI impressions where the report is available, but it still does not tell the whole story. It does not show every AI platform, and impressions alone do not tell you whether users engaged, converted or made an enquiry after landing on the site.

That is where GA4 is still useful.

If your content is being used by platforms such as ChatGPT, Perplexity, Copilot, Gemini, or Claude, you may see some of those visits appearing in your acquisition data.

How to get the data

1. Log in to Google Analytics 4
2. Navigate to Reports then Acquisition then Traffic Acquisition
3. Set the date range to the last six to twelve months
4. In the Session source / medium dimension, look for entries containing chatgpt.com, perplexity.ai, you.com, copilot.microsoft.com, bing.com, gemini.google.com, claude.ai or other AI/search sources
5. Export this data as a CSV using the download icon
6. Open the file in a plain text editor and paste it into ChatGPT with the prompt below.

Prompt 4: Analysing AI referral quality

Analyse this GA4 traffic acquisition data. Identify all sessions from AI search
platforms (ChatGPT, Perplexity, Bing Copilot, You.com, or similar). For each
AI source: show total sessions, average engagement rate, and average session
duration. Compare these metrics to my overall organic search traffic. Which
pages are receiving the most AI referral traffic?
[paste CSV data here]

What to look for

The first thing to check is whether AI platforms are sending any traffic at all. For many sites, the numbers will still be small. That does not make the exercise pointless. A small number of visits can still show which pages are being surfaced by AI tools.

It is also worth comparing GA4 with the Generative AI features report:

• A page with high Generative AI impressions but no AI referral traffic may be visible in Google’s AI results but not attracting clicks.
• A page with AI referral traffic from ChatGPT or Perplexity may be performing well outside Google’s own AI features.
• A page with standard organic traffic, no AI impressions and no AI referrals may still be doing well in conventional SEO, but not yet gaining much AI-led visibility.

If a page ranks well in Google but receives no AI referral traffic and no Generative AI impressions, it may still be a strong SEO page but not especially useful to AI systems. That could be down to structure, lack of clear answers, weak entity signals, thin supporting detail, or simply the way the topic is being handled by AI platforms.

Step 3: Use the Generative AI Report to Prioritise Pages

Before making technical changes, I would now add one extra step: use the Generative AI features report as a prioritisation layer.

In GSC, review:

1. Top pages — which URLs are getting the most generative AI impressions?
2. Dates — are impressions increasing, stable or dropping?
3. Countries — are impressions coming from the locations you actually target?
4. Devices — are mobile or desktop users seeing your content more often in AI features?

This gives you a more practical starting point than guessing from query data alone. If a page is already receiving AI impressions, Google is at least considering it relevant enough to show in generative AI features. That page may only need refinement rather than a complete rewrite.

For each high-impression page, I would then check:

• Does the page answer the likely query clearly near the top?
• Are the headings descriptive enough to stand alone?
• Is there original insight, experience, data or examples?
• Are important claims supported with credible sources?
• Is the page crawlable, indexable and eligible to show a snippet in Google Search?
• Does the page include useful images, video, tables or structured content where relevant?

Google’s own guidance is important here. Google says standard SEO best practice remains relevant for generative AI search because these features are rooted in its core Search ranking and quality systems. It also says content needs to be crawlable, indexable and eligible to show a snippet to be eligible for generative AI features [9].

That means the answer is not to create a separate “GEO” version of every page. It is to make the existing page genuinely more useful, clearer and easier to understand.

Step 4: Three Technical Changes That Can Help AI Visibility

Once you have reviewed GSC, the Generative AI features report and GA4, the next step is to improve the pages most likely to matter. There is no guaranteed formula for AI citation, but there are sensible technical and content changes that should make a page easier to understand, extract and reference.

The three areas I would usually start with are schema, AI-readable guidance files, and content structure. I would treat these as supporting measures, not shortcuts.

Action 1: Add useful schema to your highest-priority pages

Start with the pages identified in GSC as having high impressions, question-led queries, weak click-through rates or strong Generative AI visibility. If those pages already answer common questions, FAQ schema can help make that content more clearly structured.

Each FAQ item should answer a real query from your GSC data where possible. The wording does not need to be forced, but the question should be close enough to match the way people search.

<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{
"@type": "Question",
"name": "What is Generative Engine Optimisation?",
"acceptedAnswer": {
"@type": "Answer",
"text": "Generative Engine Optimisation (GEO) is the practice of optimising content to earn citations and mentions within AI-generated search responses, such as Google AI Overviews, ChatGPT Search, and Perplexity answers."
}
}
]
}
</script>

For important articles and guides, it also makes sense to use Article schema to define the author, publish date and modified date. Organization schema can help connect the content to the business behind it. The schema.org vocabulary uses American spelling for Organization, so that spelling needs to be preserved in the markup.

A small caveat is needed here. Google has said structured data is useful for eligible rich results and helping Google understand pages, but it should match visible content and should not be treated as a way to force inclusion in AI features [9]. The broader aim is to make the page clearer and more trustworthy, not to add markup for the sake of it.

Action 2: Add or update your llms.txt file

The llms.txt file is a proposed standard for giving AI systems a clean, structured overview of a website [6]. It is normally placed at the root of the domain, in the same way you would have a robots.txt file.

I would be careful about overstating this. llms.txt is not an official W3C or IETF standard, and there is mixed evidence on how consistently major AI crawlers use it. OpenAI and Perplexity both provide documentation around crawler control using robots.txt, but that is not the same thing as saying every major AI system actively relies on llms.txt [4] [5]. Google has also said Search does not use llms.txt for generative AI visibility [9].

I would treat it as a low-effort supporting file rather than a core Google SEO requirement.

# Your Site Name

> One sentence describing what your site is about and who it is for.
## Key Pages

- [Page Title](https://yourdomain.com/page-url): One sentence description
- [Page Title](https://yourdomain.com/page-url): One sentence description
## About
Two to three sentences about the author or organisation, their expertise, and why they are a credible source on the topics covered.

For larger sites, you can also consider an llms-full.txt file containing fuller markdown versions of key content. I would treat this as a supporting measure rather than a replacement for strong HTML content, schema, internal linking and crawlable pages.

Action 3: Restructure high-value content into clearer sections

AI systems tend to work better with content that is clearly structured. Dense prose can still rank, but it is not always easy for an AI system to extract a clean answer from it.

For important pages, I would look at the following:

• Use descriptive H2 and H3 headings
• Answer the main question early in each section
• Keep paragraphs reasonably short
• Use tables or lists where they genuinely help comparison
• Add examples, data and sources where appropriate
• Avoid vague claims that are not backed by anything specific
• Include useful images or video where they help the user understand the topic

If two pages cover the same topic, the more useful page is normally the one that adds something specific, e.g. original data, a better explanation, clearer examples, or practical steps based on real experience. Thin summaries are easier to produce, but they are also easier to ignore.

What I Actually Found

When I ran this analysis on my own site, the results were more striking than I expected. A single blog post on AI and the future of SEO had accumulated 416,000 impressions across 1,000 queries over 16 months — but generated only 528 clicks. That is an overall CTR of 0.068%.

The suppressed clicks prompt made the pattern immediately obvious. Here are the top queries by impression volume:

The post was ranking at position 3.88 for "ai and the future of seo", yet that single query produced 21,485 impressions and just 2 clicks. Every high-impression, near-zero-CTR query in the table is question-format or topic-based. These are exactly the queries where Google deploys an AI Overview, meaning the page was ranking well but wasn't getting the traffic.

The semantic cluster prompt added a second layer of insight. These queries were not all ones I had explicitly targeted. ChatGPT grouped them into a single intent cluster around AI disruption to search and flagged that the existing content was already covering the cluster and just needed restructuring to improve its citation likelihood within the AI Overviews.

The GSC data also revealed something else. Several of the URLs appearing in the results were not the page URL itself but anchor links: #how-is-ai-changing-seo, #ai-vs-seo-traditional-search, and #the-evolution-of-seo were each generating over 30,000 impressions with zero clicks. These are section headings being pulled directly into AI Overviews as cited sources. The content was being read and used but wasn't driving visitors back to the site. That is the gap this process helps you find and close.

Using the prompts and free tools mentioned in this guide, I could see the pages receiving the most AI referral traffic were also the ones where the content was already structured more clearly. They used specific H2 headings, direct answers, supporting detail and FAQ-style sections. Pages with dense, unbroken text performed less well from an AI referral point of view, even where they still had reasonable traditional rankings.

This has changed how I now approach new content. I am still thinking about rankings, internal links, metadata and all of the normal SEO basics, but I am also looking at whether each section can stand on its own as a useful answer.

Final Thoughts

Using GSC, GA4 and ChatGPT as outlined here will give you a much better starting point than we had before (and for free).

The Generative AI features report shows which pages are appearing in Google’s AI results. The standard GSC report gives the query and ranking context. GA4 shows whether AI platforms are already sending traffic. ChatGPT helps group the data, find patterns and turn exports into something you can actually use.

The key point is that SEO reporting needs to move beyond rankings alone. Rankings are still useful, but they do not tell the full story anymore. For many informational and research-led searches, the more useful question is whether your content is visible, trusted and cited where the answer is being generated.

References

[1] Ahrefs: Update: AI Overviews Reduce Clicks by 58%
[2] Pew Research Center: Google users are less likely to click on links when an AI summary appears in the results
[3] Gartner: Gartner Predicts Search Engine Volume Will Drop 25% by 2026
[4] OpenAI: Overview of OpenAI Crawlers
[5] Perplexity: Perplexity Crawlers
[6] llms.txt: A proposal to provide information to help LLMs use websites
[7] Google Search Central: Introducing Search Generative AI performance reports in Search Console
[8] Google Search Console Help: Generative AI performance report
[9] Google Search Central: Optimising your website for generative AI features on Google Search