Search Console AI Overviews report: what it shows and what it doesn't

AI Overviews are AI-generated summaries in Google Search. Google's launch post lists what the generative AI reports show: impressions in generative AI features on Search and Discover, the pages that appeared, countries, devices (for Search) and dates, down to hourly granularity¹. Google says the same data also stays in the overall Performance report¹, and its documentation on AI features says they are counted there under the Web search type². Google's guide to optimising for generative AI search points site owners to the same reports³. An SEO agency's news coverage of the launch also described them as impressions reporting⁴.

In plain terms: you can see that your page was shown in an AI feature. You can slice it by page, country, device and date. That is useful. A page that earns impressions in AI features is one Google's systems consider worth showing there.

Google's guide treats optimising for generative AI search as still being SEO³. So the report fits into work you probably already do. If you hear the term generative engine optimisation (GEO), read it with that in mind.

Four gaps matter.

• Impressions, not visits. An impression means your page appeared. It does not mean anyone read it. Google's launch post lists impressions, pages, countries, devices and dates; clicks and positions are not on that list, and Google says it may add metrics over time¹. Check the current help documentation before you build a dashboard on it.
• Google only. Search Console reports on Google's own surfaces. Google keeps its measurement inside its own reports, which is what walled garden means. Answers inside ChatGPT or other assistants are outside it.
• No money. The report has no revenue, margin or customer data. A page can earn impressions and bring nothing you can bank.
• No explanation. It does not tell you why a page appeared or what you should change.

Our view: treat the report as a counter, not a verdict. It tells you where to look next.

You will not get one clean number. You can build a reasonable estimate from several imperfect ones.

1. Compare by page. Take the pages with impressions in AI features. Set them against the same pages' click-through rate (CTR): clicks divided by impressions. Watch the direction over weeks, not the level on one day.
2. Watch branded search. If more people see your brand in AI answers, some will search for it by name. Trend branded queries in Search Console.
3. Check referrals in your analytics. Visits from an AI assistant appear as a referral source only when a referrer is passed, so treat this view as a floor, not a total.
4. Ask customers. A "how did you hear about us" question at checkout catches mentions no tool records.
5. Annotate changes. Mark launches, content updates and reporting changes on the timeline so you do not credit AI for a promotion.

Why bother with clicks? Pew Research tracked the Google searches of a panel of US adults and found that users were less likely to click a search result link when an AI summary appeared in the results⁵. It is a US panel and one search engine, so it says nothing direct about Europe. But it is a fair reason to look at clicks and not impressions alone.

Not first. Start with Search Console, your analytics and a short list of questions your customers ask. That costs time, not licence fees.

Our view: a paid tracker runs a fixed set of prompts and records what comes back. Its numbers describe those prompts, not the whole market, and they are not a market share. Treat them as a trend on a sample you chose.

A tracker earns its fee in one case: you need to watch several assistants, or compare yourself with named competitors, and Search Console cannot do that. If that is you, read our guide to AI visibility tracking tools before you buy.

Report a trend with caveats, not a score. A single "AI visibility score" hides the prompts, the sample and the date range.

A short format that holds up:

• What we measured: the source and the date range.
• What moved: impressions in AI features, branded queries, referrals.
• What we cannot see: other assistants, visits without referrers, revenue by AI source.
• What we propose: the next page or topic to improve, and how we will know.

Keep the same definitions each month. A report that changes its method every quarter shows nothing.