Google Search Console and AI Overviews: what you can and cannot see

What Search Console includes

When your page is cited in an AI Overview, Google counts that as an impression in the Performance report, and a click through to your site counts as a click. The activity is present in your totals.

What is absent is the label. Search Console does not separate AI Overview impressions from ordinary result impressions, so there is no filter, dimension or export that isolates them. The data is mixed at source, not hidden behind a setting.

This is worth stating clearly because a great deal of published advice implies otherwise, usually by describing a workaround that infers rather than measures.

What the numbers do tell you

Impressions and clicks by page remain useful. If a page's impressions climb while its clicks stay flat, something is showing your content without sending the reader onward, and an AI Overview is one plausible explanation among several.

Average position moves in ways that are easy to misread when overviews are involved, because the position of a citation inside an overview is not the same kind of measurement as the position of a blue link. Treat sharp unexplained changes as a prompt to look, not as a finding.

The pattern worth watching is the ratio rather than either number alone. A sustained divergence between impressions and clicks on informational pages is the clearest signal available from this data, and it is still circumstantial.

Getting the data out

The Performance report exports directly, and for anything regular the Search Console API is worth the setup. It takes a site identifier, a date range and a set of dimensions, and returns rows of clicks and impressions.

For a domain property the site identifier takes the form sc-domain: followed by your domain, which trips up nearly everyone the first time. A URL-prefix property uses the full URL instead, and mixing the two produces a permission error that reads as though access is missing.

Group by date and page. Adding the query dimension gives you visitors' own search terms, which is personal data you probably have no product need for, and declining to request it is both simpler and easier to justify.

Combining it with what you do have

Search Console tells you about Google Search. Your server logs tell you which AI crawlers read your pages. Your analytics tells you who arrived and what they did. None of the three is sufficient and the combination is genuinely informative.

A page that AI crawlers fetch often, whose Search Console impressions are rising, whose clicks are flat, and whose Direct traffic is climbing is a page being used to answer questions without being visited. That is four weak signals pointing the same way, which is the most confidence this data supports.

It is not proof and should not be reported as such. Four circumstantial signals agreeing is a strong reason to act and a poor reason to claim a number.

What to stop trying to do

Stop looking for the AI Overviews filter. It does not exist and the effort spent hunting for it is the most common waste in this area.

Stop reporting an AI Overviews impression count. Any such figure is inferred, and inferring it from a mixed total means the error bars are wider than the number.

Do report the divergence between impressions and clicks, labelled as what it is. It is honest, it is actionable, and it does not collapse the moment somebody asks how you calculated it.

What to report when the data cannot answer the question

Somebody will ask how much traffic AI Overviews sent. The honest answer is that Search Console does not separate it, and saying so plainly is better than producing an inferred figure that will not survive scrutiny.

What you can report is the divergence: impressions on informational pages rising while clicks stay flat, with the ratio shown over several months. Label it as consistent with content being used to answer questions without a visit, which is what it is.

Pair it with crawler activity from your logs for the same pages. Two independent signals moving together is a stronger argument than either alone, and it is still circumstantial, which is how it should be described.

Refusing to produce the number people asked for is uncomfortable once. Producing it and being unable to defend it is uncomfortable repeatedly, and it costs you the credibility of the figures that are solid.

Using it as context rather than as measurement

Search Console is best treated as one input among several rather than as the answer to an AI question it was never built to answer.

Its real value here is the page-level view of impressions and clicks over time, which is the longest reliable series most sites have about how their content is surfaced. That series is worth watching regardless of what portion is AI.

Combined with crawler activity and your analytics, it contributes to a picture. Read alone and asked to answer questions about AI Overviews specifically, it produces inference dressed as data.

A closing note on how this is likely to change. Google has adjusted what Search Console reports before, and the pressure to give site owners visibility into AI Overviews is real and public. If that reporting arrives, the guidance here changes and the honest answer becomes a measured one rather than an inferred one.

Until then, treat any claim of AI Overview measurement from this data as inference, including any you construct yourself.

In the meantime, the most useful posture is to keep the series clean and the definitions stable, so that if better reporting does arrive you have a comparable history to read it against rather than starting from the day the feature shipped.