Google Search Console details how to measure clicks and impressions inside AI Overviews

The Search Console performance report documentation explains, with specific rules, how clicks, impressions and position are counted inside AI Overviews and AI Mode. For those doing GEO, this is the first official measurement yardstick, even if incomplete.

For months, those working on generative search optimization (GEO) relied on inference: organic traffic dropped, it's suspected that the AI Overview "stole" the click, but there was no way to confirm this with first-party data. The Google Search Console help page about impressions, position and clicks solves part of this problem by documenting, in detail, how the performance report specifically treats AI Overviews and AI Mode as search result types.

It's not a feature announced with fanfare: it's methodology documentation, the kind of page most people ignore until they need to justify a number to a client or to management. An important caveat is worth noting before getting into the details: these counting rules don't come with a dedicated filter in the Performance report that isolates AI Overviews or AI Mode as their own segment. Today, this data falls under the generic "Web" type, mixed in with traditional links, which greatly limits the practical reading of what follows.

What the documentation confirms, by result type

The documentation details its own counting rules for each type of element:

The part that matters most for those doing GEO is the position row: an entire AI summary occupies a single value in the report, even when citing five or ten different domains. This means that the average position metric, on its own, doesn't tell you whether your link appeared first or last inside the summary, only that the summary, as a block, landed at a certain point on the page.

The limit that still exists in practice

Here lies the most important caveat for anyone trying to use this day to day: as this article has already pointed out before, there is no (yet) isolated, highlighted line saying "this came from AI Overview" inside the Performance report. Clicks and impressions generated by AI Overviews and AI Mode are still counted together under the "Web" type, without a Search appearance filter that allows separating them from traditional links.

In practice, this means that the counting rules described above work internally, shaping the numbers you see, but are not exposed as a selectable segment in the report. Anyone wanting to estimate the zero-click effect of a specific AI Overview still needs to resort to indirect proxies (a drop in aggregate CTR on queries where it's known, through other means, that an AI summary appeared), not a native filter that isolates this traffic.

The trap of aggregation by property

The Search Console documentation itself uses the example of a knowledge panel with multiple links to explain a distinction that applies equally to AI Overviews: when data is aggregated by property, a summary that cites the same domain three times generates a single impression; when aggregated by page, each URL cited within the summary counts separately.

This changes the reading of share-of-voice within an AI summary. A brand that gets cited multiple times in the same AI Overview, with different pages from the site, only reveals this citation volume when the report is segmented by page, not by domain. For anyone measuring GEO focused on how many times content is reused by AI, aggregating incorrectly under- or overestimates presence.

What AI Mode adds (and complicates)

AI Mode works as a conversational extension of AI Overviews: the user's question is split into subtopics, searched simultaneously, and each follow-up question within the same session is treated as a new, independent query in the report. In practice, this fragments the search journey into several successive "queries," each with its own set of impressions and clicks.

For growth teams, this fragmentation is valuable data, but also extra work: a single user search intent can generate multiple lines in the report, making it harder to attribute a conversion to an original question. Generative search funnel analysis therefore needs to look at sequences of related queries, not just aggregate volume per term.

What the documentation still doesn't solve

Beyond the absence of an isolated filter, the documentation makes clear that there are other important limits for those doing GEO:

• It only describes behavior for when your page has already been cited inside an AI summary; there's no visibility into how many times a summary appeared for your keywords without including your domain, which leaves the question "why wasn't I cited" without a native answer.
• Data from Search Labs experiments is left out of the report, both for AI Overviews and AI Mode, because Google treats these variations as still under active development.
• The page itself warns that the position and counting heuristic "is subject to change," which is a direct warning: dashboards and automated reports that rely on these rules need periodic review, not a one-time setup.

Why this ends part of the guesswork, but not all of it

In short: the counting documentation for AI Overviews and AI Mode gives, for the first time, official clarity about how click, impression and position behave inside generative search. This helps explain why the average position metric can be misleading, and why aggregation by property versus by page changes the outcome so much. But without a dedicated filter that isolates these segments in the Performance report, this data today remains mixed with the "Web" type, and estimating the real zero-click impact still depends on indirect reading, not an exportable, isolated number.

What this doesn't solve is full funnel visibility: without knowing the share of queries where an AI summary appears and the domain isn't cited, and without a native way to separate this traffic from the rest of organic search, GEO analysis remains incomplete, even with the counting rules documented. The logical next step for anyone measuring this seriously is to cross-reference this documentation with third-party SERP tracking, which captures the domain's presence (or absence) in summaries regardless of whether it generates a click or not.