What it is
In June 2026 Google Search Console (Google's free tool for seeing how your site performs in search) added a Generative AI performance report. For the first time, it separates impressions (the times your pages were shown) in Google's AI features from ordinary Search data. Those features include AI Overviews, the AI-written summaries at the top of some Google results, and AI Mode, Google's chat-style search.
For two years the standard complaint about AI search was that you couldn't measure it. You could watch your traffic fall and guess why. Now there's a number that comes straight from Google, free, in a tool you already have. It's limited, with impressions only and no clicks or search queries, but it's the first official look at whether Google's AI answers are using your content.
This is a walkthrough of what the report shows, what it leaves out, and how to read it. We've used this site's own data throughout, because a worked example is easier to follow than a description.
Finding it
In Search Console, open Performance, then look for the Generative AI report link under the main chart. It's marked Beta, and it shows the last three months by default.
You get impressions, meaning how often your pages appeared in AI Overviews or AI Mode, broken down by page, country, and device. You don't get clicks or queries, and the data doesn't go back very far. Use it to see which way things are moving, not as an exact count.
How to read it, with this site as the example
Over a recent three-month period this site recorded 411 impressions in Google's AI answers. That's a small number, so we looked at how those impressions were spread across the pages.
The top pages, in order, were the page that defines the site's core term (139 impressions), the main page for its second subject area (113), the homepage and the landing page for its main section (105 between them), then one blog post (31), and a long tail of reference pages such as the glossary, the playbook, and the audit checklist.
Two things jump out of that pattern.
On this site, AI answers mostly showed definition pages. The pages doing the work are the ones that answer "what is X?" for a specific term, in a paragraph that makes sense on its own. Nothing promotional shows up anywhere in the list. If you want a prediction you can test on your own data, it's this one. Look at which of your pages appear, and you'll probably find they're the ones a model could quote a sentence from without needing the rest of the page.
Of the five blog posts live then, one appeared. It was built on a named, dated study with hard numbers. The posts built on general argument didn't register at all. That's one site's data and doesn't prove anything by itself, but it's cheap to check against your own content.
The trend line matters more than the total
On this site, daily AI impressions ran between four and ten through May, June, and early July, jumped to eighteen in mid-July, then slid toward zero across August. The site hadn't published or properly updated anything since late 2025, and we think that was the likeliest cause. Every sign of freshness it gave off was stuck at the same old date. That included the timestamps in its sitemap (the file that lists a site's pages for search engines), the dates printed on its pages, and the "last updated" line, November 30, 2025, in three plain-text files written for bots: robots.txt (the file that tells bots which pages they may visit), llms.txt (a list of key pages some sites publish for AI tools), and ai.txt (a statement of how AI tools may use the content).
Ahrefs, which makes SEO tools, studied about 17 million AI citations and found that pages cited by AI assistants were on average about 25.7% newer than pages in Google's ordinary results. Google's AI Overviews were the exception, citing pages closer in age to the ordinary results. So freshness may matter less for Google's AI answers than for other AI tools, and our explanation for the slide is a likely cause, not a proven one.
If your Generative AI line is drifting down while your ordinary Search line holds steady, stale content is still the cheapest thing to check first.
What the report cannot tell you
Be clear about the limits before you build any reporting on it.
- No queries. You can't see what someone asked that brought up your page, so you can't do keyword research with it.
- No clicks. Impressions aren't visits, and a page can be shown often in AI answers without many people clicking through to it.
- Google only. ChatGPT, Perplexity, Claude, and Copilot don't show up here at all, and what AI tools cite often isn't what ranks in ordinary search. Ahrefs found only about 11% of the pages AI tools cite also rank in Google's or Bing's top 10 for the same question. The AI tools don't agree with each other either: Profound ran 100,000 questions through both ChatGPT and Perplexity and found only 11% of the websites they cited were cited by both. A measure that only covers Google can't tell you how visible you are in AI answers overall.
- Your analytics can miss AI visits too. Some visits from AI tools arrive without a referrer (the note a browser passes on about which site a visitor came from), so analytics files them under "Direct", and GA4 (Google Analytics 4) can undercount AI traffic as a result.
The opt-out setting Google shipped alongside it
On the same day as the report, Google announced a toggle in Search Console that keeps your site out of its generative AI features in Search, such as AI Overviews, AI Mode, and AI Overviews in Discover. Google says the setting "will not be used as a ranking signal for search results outside of these generative AI Search features". Sites that switch it on also lose any traffic and impressions from those features.
Until this toggle, those features simply followed the rules you set for
Googlebot (the bot Google sends to read pages for search). There's also a
separate name you can use in robots.txt, Google-Extended, but it does
something else. According to
Google's documentation on its bots,
it controls whether Google can use your content to train its Gemini models,
and to supply facts to Gemini's answers in the Gemini app and in Vertex AI
(Google's AI platform for developers). Google says it doesn't affect your
place in Google Search, so it never covered AI Overviews or AI Mode.
Whether to use the new toggle is a business decision more than a technical one. Blocking protects you against substitution, where an AI answer gives the user everything they need and no visit follows. Allowing keeps you in the running to be cited, along with the brand exposure that comes with it. SparkToro, an audience research company, used Similarweb data and found that about 68% of US Google searches in the first four months of 2026 ended without a click. That substitution happens to cited and uncited sites alike.
Most publishers should leave it open and compete on being the best source. Sites where the information itself is the whole product, where the answer is the business, have a strong case for blocking.
What to do with this
A monthly routine
- Check the Generative AI report once a month and watch the trend, not the total.
- List which of your pages appear. Write more pages like those, and open each one with a definition that can be quoted on its own.
- If the line is sliding, check your freshness signals before anything else: true published and updated dates in your page code, an accurate sitemap, and content that has actually been revised.
- Pair it with something that sees the other AI tools. Checking your server logs (your host's record of every request) for visits from AI crawlers (the bots AI companies send to fetch pages) is free and shows who is fetching your pages. Paid AI-visibility tools show where you appear in answers.
- Don't set a target on impressions alone. It covers one search engine and one measure, with no clicks.
References
- Google Search Central, "Introducing generative AI performance reports", June 3, 2026
- Google, "New opportunities, control and insights for website owners", June 3, 2026, updated August 31, 2026
- Google, "Google's common crawlers", Google crawling documentation
- Ahrefs, "AI search overlap", August 11, 2025
- Profound, "Answer Engine Citation Overlap Strategy", July 1, 2025
- Ahrefs, study of how fresh AI-cited pages are, July 28, 2025
- SparkToro, "In 2026, less than one third of Google searches still send a click", June 8, 2026