AI Overview Visibility: What Search Console Now Shows
Google's generative AI report reached every site on August 31. How to read it, what it leaves out, and how to check the AI search it will never cover.
For about eighteen months, the only honest answer to “are we showing up in AI Overviews?” was a shrug and a spreadsheet of manual searches. That changed on August 31, 2026, when Google finished rolling out its Search Console generative AI performance reports to every site worldwide, after a phased test that started on June 3. Open Search Console today and you can see how often your pages appeared in AI Overviews and AI Mode.
How often is the whole of it. The report carries no clicks and no queries, and it covers Google’s own search surfaces only, which on current usage numbers is a minority of the AI search your buyers are doing. I still think it’s the most useful thing to land in Search Console in years, as long as you know what you are holding.
What the generative AI report actually contains
The generative AI performance report shows impressions from AI Overviews and AI Mode on Google Search, broken out by page, country, device and date. There is a separate version for Discover. An impression means a link to your site was shown to someone inside a generative AI feature.
The exclusions are what I’d read first:
- No clicks. Google’s documentation lists impressions and nothing else. You cannot calculate a click-through rate for AI Overviews from this report.
- No queries. The report will tell you a page was surfaced 4,000 times last month. It will not tell you what people asked to make that happen.
- No Search Labs data. Experimental features still in Labs are left out entirely, so anything Google is testing this quarter is invisible until it graduates.
One detail catches people out. This data was already inside your existing Web performance report. Google is filtering numbers you have had all along, not adding new ones, so your site-wide impression totals did not jump on August 31. If yours did, something else changed.
The missing click column, and what you can still do with impressions
Impressions without clicks sounds useless. It isn’t, because the interesting number was never the absolute count. It’s the ratio between a page’s AI impressions and its ordinary Web impressions, tracked over months.
Here is how I read it. Export both figures per page and put them side by side:
| Pattern | What it usually means |
|---|---|
| AI impressions rising, Web impressions flat | The page is being pulled into answers for queries it never ranked well for |
| Both rising together | Normal growth. Nothing AI-specific to investigate |
| AI impressions rising, Web clicks falling | The page is being summarised instead of visited |
| Almost no AI impressions on a strong page | A retrieval problem, not a ranking problem |
That last row is the one worth acting on. A page that ranks in the top five and picks up close to zero AI impressions is usually structured in a way that makes it hard to quote, and the fix is passage-level: lead each section with the answer, keep claims next to their evidence, restate the subject inside the section rather than only in the heading. That work is the whole of how to get cited by AI search engines, and this report is the first way to check whether it worked without running fifty manual prompts.
The row above it is the uncomfortable one. Rising AI impressions against falling Web clicks is your content doing its job and your analytics recording nothing, which is a measurement problem rather than a content problem. I’ve written about what that does to attribution separately, because it outgrew a section.
The opt-out toggle, and why I would leave it alone
The same rollout added a control letting you block your content from appearing in and grounding Google’s generative AI features. Search Engine Land’s coverage of the global rollout notes that Google says the setting is not used as a ranking signal for regular search results, so flipping it should not hurt your blue-link visibility.
I would still leave it alone for almost every commercial site. Turning it off removes you from AI Overviews and AI Mode without removing the AI Overview from the query. The answer still gets generated. Your competitors still get cited in it. You have opted out of the citation, not out of the format.
The exception is publishers whose business model is the pageview itself, where a citation that satisfies the reader is a direct revenue loss rather than an awareness gain. If that’s you, the calculation is genuinely different and worth running properly. For everyone selling something at the end of a research process, being quoted in the answer is the point.
Google is a fraction of the AI search you should care about
Now the part the launch coverage mostly skipped. Search Console’s report covers AI Overviews and AI Mode. It does not cover ChatGPT, Perplexity, Claude, Microsoft Copilot, or the Gemini app. Those are separate products with separate indexes, and the zero-click problem lives in all of them.
The scale is not marginal. StatCounter’s measure of AI chatbot usage worldwide put ChatGPT at 79.4% in August 2026, with Gemini at 10.9%, Perplexity at 4.31%, Copilot at 2.79% and Claude at 2.57%. Similarweb’s measurement of generative AI web traffic disagrees sharply, putting ChatGPT nearer 53% by May 2026 as Gemini climbed toward 27%. Different methodologies, very different numbers, and I would not build a forecast on either. What both agree on is the shape: a large majority of assistant usage happens somewhere Google’s new report cannot see.
So after August 31, Google has closed one blind spot and left the bigger one open. Treating the Search Console number as your AI visibility metric would be a mistake, and an easy one to make now that there’s finally a chart to screenshot.
Checking the four surfaces the report misses
Three of these are free, and one of them is better than what Google shipped.
Bing Webmaster Tools, for Copilot. Microsoft got here first and went further. AI Performance went into public preview in February 2026 and reports actual citations across Copilot and Bing’s AI summaries, plus grounding queries, the phrases the model generates internally when it goes looking for a source. Citations and queries are exactly the two things Google withholds. If I could only verify one property this month, it would be this one. The grounding queries alone will change what you think your pages are about, and my Bing Copilot citations guide covers the monthly loop that turns them into edits.
Server logs, for ChatGPT. OpenAI runs
three separate crawlers and the distinction
is the useful bit. OAI-SearchBot surfaces sites in ChatGPT’s search features,
ChatGPT-User fetches a page because someone asked for it, and GPTBot collects training
data. This is the one check here you’ll likely need a developer for: ask for a monthly
count of the first two per URL from the server logs, which is a ten-minute request. That
count is a rough read on ChatGPT retrieving your pages, and no dashboard offers it. It’s
crude and it undercounts cached answers, and it is still the only first-party signal
available.
Referral traffic, for everything. Segment sessions from chatgpt.com, perplexity.ai,
claude.ai, copilot.microsoft.com and gemini.google.com in GA4. Volumes are small.
Read them as a trend line and an intent signal rather than a traffic channel, and check
which pages they land on, because that tells you which of your pages the assistants are
handing people. For Perplexity specifically, my citations guide
covers the crawler settings that silently keep sites out of its answers.
Manual prompt testing, for the truth. Pick the fifteen questions your content exists to answer. Ask them in ChatGPT, Perplexity, Gemini and Copilot on the same day each month. Record who gets cited. It takes an hour, it is the only direct read on citation share, and you cannot backfill it, which is the argument for starting this month rather than next quarter.
A monthly routine that takes about an hour
The reports are new. The habit of reading them is what produces anything useful, so here’s the version I’d run.
- Export AI impressions and Web impressions per page from Search Console. Same date range, month over month.
- Flag every page where the two lines diverge in either direction. Divergence is the signal. Growth in both is just growth.
- Pull citations and grounding queries from Bing Webmaster Tools. Compare the grounding queries against the keyword you actually targeted.
- Run the fifteen prompts. Log which domains were cited, not only whether you were.
- Fix one page. Rewrite the sections of the strongest divergent page to be quotable on their own, and note the date so next month’s export means something.
Step five is where the value is. The first four steps produce a dashboard, and a dashboard that never changes a page is a hobby.
The short version
Search Console now shows AI Overview and AI Mode impressions for every site, with no clicks and no queries, and that data was already sitting inside your Web performance report in aggregate. Read it as a ratio against ordinary impressions rather than as a number. Leave the opt-out toggle alone unless pageviews are your product. Then check Bing, your server logs and the assistants themselves, because Google measuring Google tells you about a minority of the AI search happening around your category. I don’t expect that gap to close, so I’ve stopped waiting for one report to cover it.
If this sits inside a wider remit, the AI guide for SEO specialists covers where visibility auditing fits against the rest of the week, and the rest of the AI SEO and GEO guides go deeper on earning the citation once you can see whether you have one. For how search fits an AI marketing practice overall, start with the complete guide to AI marketing.