Google’s generative AI performance report is the first first-party number for how often a site appears in AI Overviews and AI Mode. It reports impressions: how many times links to your pages were shown in those features. It does not report clicks, queries, position, or the answer your link appeared in. Read as exactly that, it is useful. Read as “how much Google’s AI uses us,” it is wrong.
This post covers what the report contains, how its impressions are counted and aggregated, what it leaves out, the new control that can make it go blank, and four ways to use it that stay within what it measures. It belongs to the measurement stack in Measuring AI visibility without overstating it and sits under How to audit a site for AEO and GEO.
What is the generative AI performance report?
It is a pair of Search Console reports that isolate impressions from Google’s generative AI features. Google announced them on 3 June 2026, rolled them out first to a subset of sites, and added a note that “as of August 31, 2026, we’ve rolled out these insights to all websites worldwide” (Google Search Central Blog).
There are two reports:
- Search covers AI Overviews and AI Mode (Search Console Help).
- Discover covers generative AI features in Google Discover (Search Console Help).
The data is not new data hidden until now. Google’s announcement says it “is included in the overall performance report, where it will continue to be tracked.” The new reports are a separate view of impressions that were already inside your totals.
What does it show?
It shows one metric, impressions, cut by four dimensions. The Search report’s help page lists them:
| Dimension | What it groups by | Note |
|---|---|---|
| Pages | ”The final URL linked by a generative AI feature after any redirects” | Mostly assigned to the canonical URL, as in the Performance report |
| Countries | Where the search originated | |
| Devices | Desktop, tablet, or mobile | Search report only |
| Dates | Day, week, or month, in Pacific Time | Newest data can be preliminary |
A search-type filter separates “Web: text-based” from “Web: multimodal,” which covers searches started from images, including Google Lens and Circle to Search. Both chart and table export, with ”~” and ”-” values written as zeros in the download.
An impression here is “how many times links to your site were shown to a user in a generative AI feature on Google Search.” Standard impression rules apply. In the Performance report’s methodology, an AI Overview “occupies a single position in search results, and all links in the AI Overview are assigned that same position” (Search Console Help).
How are the impressions aggregated?
The chart and the table count differently, and the gap between them is a known property, not an error. The chart is aggregated by property: “if two results from the same site appeared in a generative AI search results feature, they count as a single impression in the chart total.” The table, grouped by page, counts each URL separately (Search Console Help).
So an AI Overview that links three of your pages contributes one impression to the chart and three rows of one impression each to the page table. Sum the page table and it will exceed the chart. Report the chart total as “AI feature appearances” and the page table as “page appearances,” and don’t add one to the other.
Discover works differently. Its report is aggregated by page throughout, and an impression is counted only when the card is “scrolled into view,” once per result per session (Search Console Help). Don’t add Search and Discover AI impressions together either; they count different events.
What does it leave out?
It leaves out everything that would turn an impression into an outcome. In particular:
- Clicks. The report has no click metric. Clicks from AI Overviews and AI Mode are still counted in the main Performance report, under the “Web” search type (Google Search Central), but there they are mixed with every other web result and can’t be separated.
- Queries. There is no query dimension. You can see which pages appeared, not which questions triggered them.
- The answer. An impression says a link was shown. It does not say what the answer said, whether your page was the source of a claim, or where in a long AI Mode response the link sat.
- Grounding without a link. Google’s control documentation distinguishes content “showing up as links” from content “helping to ground AI responses” (Search Console Help). The report counts links shown. Use that does not produce a visible link is outside it.
- Search Labs. “Search Console doesn’t include data from experiments in Search Labs.”
- The long tail. The Performance report’s usual limits apply, including a 1,000-row table.
None of these are criticisms of the report. They describe its scope, and the scope is what goes in your write-up.
Why might the report be empty?
The report can be empty because the site has too few AI feature impressions, or because someone excluded it. Google’s help page says a missing report “usually happens when your site hasn’t received enough impressions,” and adds that it “may be because you’ve excluded your site from Search generative AI features.”
That exclusion is a new control under Settings in Search Console, also rolled out to all sites on 31 August 2026. It has three values: include (the default), exclude, and inherit from a parent property. Excluding means “links to your site and your site’s content won’t appear” in AI Overviews, AI Mode, or Discover’s generative features, and the change “generally takes a few days.” Google states that the control “isn’t used as a ranking or inclusion signal affecting other parts of Search,” and that it “doesn’t affect AI training,” for which it points to Google-Extended (Search Console Help). Training vs. retrieval crawlers: the block decision covers that second decision.
Because child properties inherit by default, check the control on the domain property before concluding a section of the site gets no AI impressions. A URL-prefix property for /blog/ can be excluded by a setting someone made on the domain.
How should you use it?
Use it for four things, each of which stays inside what it measures.
1. Find the pages that appear, and check they are the ones you want. Export the page table for the last three months. Sort by impressions. If the top rows are a glossary entry and an old press release while your product documentation is absent, that is a finding about what Google’s AI features retrieve from you. It is worth knowing even though it doesn’t say why.
2. Compute an AI share per page. Export the Performance report’s page table for the same dates, Web search type. Join the two exports on URL in a spreadsheet. With AI impressions in B and all web impressions in C, =IF(C2=0, "", B2/C2) gives the share of each page’s impressions that came from AI features. A page with strong web impressions and an AI share near zero is worth a look. Check its snippet eligibility first, because Google requires a page to be “eligible to be shown in Google Search with a snippet” to appear as a supporting link. nosnippet, max-snippet and data-nosnippet on AI surfaces covers the directives that remove that eligibility, and AI Overviews eligibility covers the rest.
3. Join it to a readiness crawl. Add your crawler’s per-URL export as a third tab and look up each page’s status, canonical, and readiness scores. Pages with AI impressions that redirect, or whose canonical points elsewhere, are reported against the final or canonical URL, not the one you linked. Status codes, redirects, and canonical drift on AI surfaces explains why the URL in the report may not be the URL you expected.
4. Watch a before-and-after, cautiously. After a set of fixes, compare AI impressions for the changed pages against an unchanged group over the same weeks. Movement in the changed group alone is suggestive. Movement in both is the engine, not you. Crawl, fix, re-crawl covers proving the fix landed. The report can then show whether exposure moved afterwards, never that the fix caused it.
What not to do: don’t divide clicks from the Performance report by AI impressions to get an “AI CTR.” The clicks include every web result and the impressions don’t, so the ratio has no meaning.
How does it fit with other measures?
It is the best available measure of exposure on Google’s AI features, and only that. It says nothing about ChatGPT, Claude, Perplexity, or Copilot, and it shows exposure, not citation or influence. Put it in a report as its own line, named, alongside a readiness line you control and referral and citation-sample lines you don’t. What prompt tracking can and can’t tell you covers the sampling layer it is most often confused with.
GEOBee does not read Search Console and does not measure impressions. Its part is the readiness tab in step 3. A crawl exports every URL’s status, indexability, and AEO, GEO, and SEO scores as CSV (exporting data), ready to join to the report’s page export on URL.
To build the readiness tab for that join, download GEOBee and export a crawl of the pages the report lists.