Search Console has a report for generative AI. It counts how often links to your site were shown inside Google’s AI features, and it is available in the same place as the performance report you already use.
After two years of guessing whether AI Overviews were doing anything at all, a first-party number is worth having. It is also worth reading the documentation closely, because the useful part is the list of what the report contains, and the list is shorter than most people expect.
What the report counts
Google defines the metric in one sentence: “Impressions are how many times links to your site were shown to a user in a generative AI feature on Google Search.”
Two generative AI capabilities are covered: AI Overviews and AI Mode. You can break the data down by four dimensions, and that is the whole set: Pages, Countries, Dates, and Devices.
There are a few mechanics in the documentation that matter more than they look. Data is assigned to the page’s canonical URL rather than to a duplicate, so a site with messy canonicals will see its impressions pooled somewhere it did not expect. All dates run in Pacific Time, which shifts day boundaries for most of the world. Google Discover has its own separate report, so Discover impressions are not in this one. And Search Console excludes data from experiments in Search Labs, which means anything Google is still testing on users is invisible here by design.
The documentation does not mention clicks
Read the page looking for clicks or click-through rate and you will not find either word. Not in the metric definition, not in the dimension list, not in the section on reading the chart.
That is an absence rather than a statement, and the difference matters. Google does not say that clicks in AI features are not counted, or that they never will be. The documentation simply does not list them. Anyone telling you that Google has confirmed AI Overviews produce no measurable clicks is reading a claim into a gap.
What you can say with confidence is narrower and more useful: as documented today, this report answers how often you were shown, and it does not answer what happened next.
The list of surfaces is explicitly temporary
Google adds one line about the two covered features that is easy to skim past: “We expect to update this list over time as we develop Google Search.”
So the report is a snapshot of a moving target. The two surfaces named today are the two that exist in the report today, not a description of where AI answers will appear next quarter. If you build a reporting template around exactly these two, expect to rebuild it.
Why one report cannot answer the real question
The definition contains its own boundary: impressions in a generative AI feature “on Google Search.”
That is not a limitation Google is hiding. It is simply what the product is. But it means the report cannot see the places where a growing share of research now happens. When someone asks ChatGPT which vendor to shortlist, or reads an answer in Perplexity, or gets a recommendation from an assistant that never touches Google, none of it appears in Search Console. There is no row for it, because it did not happen on Google Search.
This creates an odd situation for anyone reporting to a board. You now have a credible, first-party impression count for one slice of AI search, and nothing at all for the rest. The number is right. The problem is that it is precise, official, and easy to mistake for the whole picture.
The second gap is the one the report itself points at. Impressions tell you that you were shown. They do not tell you whether you were recommended, described accurately, cited as the source, or mentioned alongside a competitor who got the better sentence. Two brands can have identical impression counts in AI Overviews while one is the answer and the other is a footnote.
What to do with the report now
Turn it on and take a baseline. Even an incomplete number is better than the estimates most teams have been working from, and you will want the history later.
Check your canonicals before you trust the page-level breakdown. If impressions are being credited to a canonical you were not expecting, that is a data-hygiene problem worth fixing before anyone builds a dashboard on it.
Set the date handling straight with whoever reads the report. Pacific Time will not line up with your other reporting, and the first person to notice a mismatch will assume the data is broken.
Then decide what to do about the part the report does not cover. If your buyers research inside assistants other than Google Search, Search Console will never show you that, no matter how the report evolves. Knowing whether you are recommended there, and in what terms, takes a separate measurement running alongside this one.
That is the honest shape of it. Google just gave you a real number for a real surface. It is one surface.
Sources
- Google Search Console Help, “Generative AI performance report”: https://support.google.com/webmasters/answer/16984139 (retrieved 23 August 2026). Source of the impression definition, the list of covered features, the four dimensions, the canonical-URL and Pacific Time notes, the Search Labs exclusion, and the statement that the feature list will be updated over time.
