Search Console's Generative AI Report: How to Read Your AI Overview Impressions

For two years, site owners could only guess how often Google's AI Overviews and AI Mode showed their pages. That changed on 31 August 2026, when Google rolled out the generative AI performance report in Search Console to every website worldwide. It is the first first-party measure of AI search visibility on Google, and it comes with limits that are easy to misread.

This guide explains what the report counts, what it leaves out, how to read it next to your normal performance data, and a simple monthly routine for agencies and in-house teams.

Related: how AI engines choose their sources

What Google launched

Google's announcement on 3 June 2026 introduced "Search Generative AI performance reports", with dedicated views for Search and for Discover. Google rolled them out to a subset of sites first, then to every site on 31 August 2026. The reports sit in Search Console next to the existing Performance report.

Google is clear that this is a separate view, not separate data. Impressions from AI features were already counted in the overall Performance report. The new report isolates them so you can see how your visibility inside AI answers changes over time.

What the report counts

According to the report's help page, impressions are "how many times links to your site were shown to a user in a generative AI feature on Google Search". Two features are included today:

• AI Overviews, the AI summaries at the top of a normal results page.
• AI Mode, Google's conversational AI search.

Google says it expects to update the list over time. Search Labs experiments are not included, because they are still in development.

The four dimensions

Search type filter

You can filter between Web: text-based (typed queries) and Web: multimodal (searches made with images, such as Google Lens, Circle to Search or Chrome image search). If your products are often photographed and searched, the multimodal view may surprise you.

How impressions are added up

The chart totals are counted by property. If two of your links appear in the same AI answer, the chart counts one impression. The table counts by page when you group by page, so the table can add up to more than the chart. Google warns about this in the help page; it is the same behaviour as the normal Performance report.

What the report does not tell you

The gaps matter more than the features, because they decide what you can and cannot conclude.

• No clicks. You see how often you were shown, not how often people clicked. To estimate AI referral clicks, compare landing-page sessions in your analytics, as described in measuring AI referral traffic.
• No queries. You cannot see which questions triggered the AI answer. You have to infer topics from the pages that earn impressions.
• No position or prominence. A link in the first line of an AI Overview and one hidden behind "show more" both count as an impression, as far as the report tells you.
• No other AI engines. ChatGPT, Claude, Perplexity, Copilot and Gemini app answers are outside Search Console.
• The usual limits apply. The 1,000-row table limit and the normal data retention of the Performance report also apply here.

Why the report might be empty

If you open the report and see nothing, Google lists two causes.

1. Not enough impressions yet. Small or new sites may not have appeared in AI features often enough.
2. You are excluded. Google's new Search generative AI control lets a property opt out of AI Overviews, AI Mode and Discover AI features. If your property, or a parent property it inherits from, is set to exclude, there is nothing to count. Check Settings, then Search generative AI. We explain the control in the Search generative AI control vs Google-Extended.

There is a third cause the help page does not mention, because it is outside Search Console: Google cannot use what it cannot crawl. If Googlebot is blocked by robots.txt, a firewall or a CDN rule, your pages cannot appear in AI features at all. Run AI Crawler Check on a key page to rule this out; it takes a few seconds.

Note that Google-Extended is not part of this. Blocking Google-Extended limits training of Google's models, but Google's help page for the new control says it is the control, not Google-Extended, that decides whether you appear in Search AI features. Google-Extended vs Googlebot covers the difference.

How to read it: four useful views

1. AI share of impressions

Open the normal Performance report and the AI report for the same date range and search type. Divide AI impressions by total impressions. That ratio is your AI exposure. A rising share means AI features are taking a bigger part of how Google shows you, whether or not total impressions change.

2. The AI-favoured pages

Group by page and export. Compare the list with your top pages by total impressions. Pages that rank high in the AI list but low in the normal list are being chosen for answers more than for classic results. Study them: they are usually short, direct, well-structured answers with clear facts. Our AI Overviews optimization guide lists the patterns.

3. Pages with impressions but falling clicks

Take pages with strong AI impressions, then look at their clicks in the normal Performance report over the same months. If impressions hold and clicks fall, AI answers may be satisfying the query without a visit. That is the zero-click problem, and the response is usually to give people a reason to click: tools, data, examples or depth the summary cannot hold. See our zero-click survival guide.

4. Country and device mix

AI Mode and AI Overviews roll out unevenly. If one country shows a sharp rise in AI impressions, check whether a feature launched there. If mobile dominates, make sure your mobile pages load fast and put the answer near the top.

A worked example

The figures below are illustrative, not from a real site, but the steps are exactly what you would do with your own export.

Read it row by row. The two how-to pages are mostly being shown inside AI answers, and their clicks are falling. That is the pattern to act on: the answer is being given in the AI feature, so the page needs something the summary cannot carry, such as a sizing calculator, a fault-code table, a short video or a clear "book an engineer" step. The product and location pages are barely in AI answers and their clicks are stable, which is normal for transactional pages. They do not need AI work first.

Notice what the table cannot tell you: which questions led to the 7,800 AI impressions on the pressure page. You would test that by asking AI Mode a few likely questions yourself and checking whether your page is cited, and how.

When AI impressions drop

A sudden fall in AI impressions usually has one of five causes. Work through them in this order, because the first two are the cheapest to rule out.

1. A crawl block. A new firewall rule, CDN setting or robots.txt change stopped Googlebot. Check with AI Crawler Check and your server logs.
2. A control change. Someone set the property, or a parent property, to exclude. Check Settings, then Search generative AI.
3. A Google update. Compare the date with the Google Search Status Dashboard. Core and spam updates move AI visibility as well as rankings.
4. A feature change. Google changes when AI Overviews appear, often without notice. If total impressions are flat and only AI impressions fell, this is likely.
5. A content change. A rewrite that buried the direct answer, removed a table or changed a canonical can drop a page out of AI answers.

Reporting it to clients and stakeholders

AI impressions are a new number, and new numbers get misread. Three habits keep reports honest.

• Name it precisely. Call it "Google AI Overview and AI Mode impressions", never "AI traffic".
• Show the share, not only the count. A rising count can simply follow rising total impressions. The share shows whether AI features matter more.
• Pair it with an outcome. Put AI impressions next to clicks, leads or sales for the same pages, so nobody mistakes visibility for revenue.

If you manage many sites, check crawl access across all of them first with our batch checker, so an access problem on one client does not distort a portfolio report.

A monthly routine

This is the routine we suggest to agencies and in-house teams. It takes about 30 minutes per site.

• □Confirm crawl access first: run AI Crawler Check and the AI crawler diagnostic so a block does not masquerade as a ranking problem.
• □Check Settings, then Search generative AI, and record whether the property includes, excludes or inherits.
• □Export the AI report by page and by country for the last full month.
• □Calculate AI share of impressions against the normal Performance report.
• □List the top 20 AI pages and note what they have in common.
• □Flag pages where AI impressions are steady but clicks fell, and plan one improvement each.
• □Record the numbers in your AI visibility KPIs dashboard alongside AI referral sessions from analytics.

What it means for GEO work

For anyone doing generative engine optimisation, the report changes what you can prove. Before August, the case for AI-focused content rested on manual prompt tests and third-party estimates. Now you can show, per page, whether Google chose your content for its AI answers, and how that changed after you rewrote a page. Use it as a before-and-after measure: record AI impressions for a page, make one focused change such as adding a direct answer in the first paragraph or a comparison table, and compare the next four weeks with the previous four. One change at a time keeps the result readable. Pages that gain AI impressions after a structural change tell you which patterns Google prefers on your site, and that is worth more than any general checklist.

Exporting the data

The report has the usual export button for the chart and the table. Values that show as ~ or - in the interface (not available) become zeros in the download, so treat zeros in an export with care. If you report to clients, label the metric clearly as "Google AI impressions" so nobody reads it as clicks or as data from other AI engines.

How it fits with the rest of your AI visibility data

No single report answers every question. The Search Console report is the only first-party number for Google, which makes it the anchor; everything else adds context.