What the report actually contains
Search Console now carries a generative AI performance report covering two surfaces: AI Overviews, the summary block that appears above conventional results, and AI Mode, the conversational search experience.
The metric is impressions, and Google defines it precisely: “Impressions are how many times links to your site were shown to a user in a generative AI feature on Google Search.” The unit is a link appearing, not a page being read.
The data can be grouped by Pages, Countries, Devices or Dates, and filtered by URL, which is the same dimensional structure as the conventional performance report.
Standard Search Console constraints carry over unchanged rather than being relaxed for the new surface. Google notes that “the usual data limitations (1,000 row limitation, time period, etc) for the Search performance report also apply to this report”, and that the newest data can be preliminary, meaning it is still being collected and may change within hours.
Search Labs experiments are excluded entirely, so anything Google is currently testing with a subset of users does not appear anywhere in the figures. All dates are reported in Pacific Time, which matters when reconciling against analytics configured to a different zone.
The metric that is missing
The report does not include clicks. It does not include click-through rate. It does not include the queries or prompts that triggered an appearance. Each of those omissions is individually defensible; together they remove almost everything a publisher would use the data for.
Count of metrics and of grouping dimensions documented for each report, read October 10, 2026
Source: Google Search Console Help, Performance report (Search results) and Generative AI performance report (Search).
The Performance report groups by queries, pages, countries, devices, search appearance and dates; the generative AI report by pages, countries, devices and dates.
For most publishers that removes the only question worth asking. Knowing that a link appeared in an AI Overview two hundred thousand times tells you nothing about whether anybody arrived, and the entire anxiety about AI answers concerns exactly that gap.
This is a measurement problem the industry has met before, and this desk has covered its shape twice: once when Google changed what an impression means in ad serving, and once when advertisers gave up on a single audience number altogether. In both cases the pattern was the same — the available metric drifted away from the commercial question.
An impression in a generative answer is a genuinely different object from an impression in a list of blue links, and treating the two as the same unit is the error the report invites. A citation sitting inside a synthesised paragraph may be read closely, glanced at, or ignored completely, and the surface is explicitly designed to satisfy the query without requiring a visit at all.
Reporting the appearance while withholding the outcome means the one number publishers can now see is the number least connected to anything they can bank. Exposure without arrival is not a business metric, and presenting it as evidence of AI performance would mislead anyone reading the deck.
Where the data already was
The most consequential sentence in the entire documentation is easy to miss, because it reads like a routine note about scope rather than a substantive disclosure. Google states that the generative AI report “includes data from the Web search type in the Performance report”.
Read plainly, that means AI impressions were never held in a separate universe. They have been inside the headline Search Console totals all along, and the new report surfaces a subset rather than adding a previously hidden stream.
That reframes a year of industry speculation about hidden AI traffic. Publishers watching their impressions hold steady while clicks fell away were not missing a parallel dataset held back by Google; they were looking at the blended figure the whole time, without any means of decomposing it into its parts.
Google's developer guidance on AI features, last updated 10 December 2025, says the same thing in different words: AI feature traffic “are included in the overall search traffic in Search Console” and appears under the Web search type.
So the correct description of the change is disaggregation rather than new disclosure, and the distinction is not pedantic. Being able to see how much of an existing total is generative is genuinely useful to anyone planning content, but it is a materially smaller claim than most of the announcement coverage implied, and it should be described accurately to clients.
What the controls cost
Google documents the levers a publisher has over appearance in AI features, and they are the familiar ones: nosnippet, data-nosnippet and max-snippet directives, noindex, robots.txt rules for Googlebot, and the Google-Extended crawler token for training use in other Google systems.
It also states that there are “no additional technical requirements” to be eligible — appearing in AI features follows from ordinary Search eligibility rather than from any opt-in.
The difficulty is that the snippet controls are blunt. Suppressing a snippet to keep content out of a generative answer also degrades how the page presents in conventional results, because the same directive governs both.
That is the trade this desk examined in our report on blocking crawlers and search visibility: the mechanisms available to refuse AI use are mostly the same mechanisms that deliver search visibility.
Notably, the documentation we retrieved does not address whether opting out of AI features affects ranking in ordinary web search. That silence is worth noting rather than filling.
How to read the number honestly
Treat the impression count as a measure of exposure surface rather than of audience. It answers one question well — is Google actually using my pages in generative answers, and which pages — and that question was genuinely unanswerable from the publisher's side until this shipped.
Do not compute a click-through rate against it under any circumstances. The clicks in the conventional report already include AI-driven visits mixed in with everything else, so dividing those clicks by generative impressions puts two incompatible populations either side of the line and produces a ratio with no defensible meaning whatsoever.
Watch the trend rather than the level. The 1,000-row cap truncates the tail and the preliminary flag on recent data makes the last few days unstable, so absolute figures are unreliable at the margins, but direction measured over several weeks is informative and comparable.
Segment by page. If generative impressions concentrate on explanatory or definitional pages while conventional impressions hold on transactional ones, that is a content-mix signal you can act on.
And be sceptical of vendor dashboards quoting AI visibility percentages. If Google does not publish clicks or queries for these surfaces, a third party claiming precise AI referral share is modelling, not measuring — a distinction we have drawn before about tools selling visibility scores.
What to watch
Whether clicks arrive at all. Google has indicated that more metrics will follow without committing to any timeline, and click data is the single addition that would turn this from an interesting diagnostic into a commercially meaningful report.
Whether query data ever appears. Prompts in AI Mode are longer and more specific than conventional queries, and publishing them would expose a great deal about how the surface actually works.
Whether the blend is ever broken out in reverse — that is, whether the conventional performance report gains the ability to exclude generative impressions, which would let publishers see the classic web number cleanly for the first time since the surfaces merged.
And whether independent measurement emerges. The absence of clicks creates a vacuum that vendors will fill with panel estimates, and those estimates will be quoted as fact unless their method is published.
Until then, the honest summary for any client or board is that appearance is now measurable, arrival is not, and anyone presenting an AI referral figure should be asked where it came from.
Methodology and limitations
The description of the generative AI performance report — the surfaces covered, the definition of an impression, the available groupings, the 1,000-row and preliminary-data limitations, the exclusion of Search Labs experiments and the statement that the report includes data from the Web search type in the Performance report — is quoted from Google's Search Console help documentation, retrieved 26 September 2026. The publisher controls for AI features and the statement that there are no additional technical requirements are quoted from Google's Search Central developer documentation on AI features, which carried a last-updated date of 10 December 2025. We attempted to retrieve Google's Search Central blog announcement of the report and it returned only navigation rather than article text, so the launch date of 3 June 2026 and the worldwide rollout by 31 August 2026 are attributed to trade coverage rather than quoted from Google. We have not tested the report against a live property.
Conclusion
Google has given publishers a real number and withheld the one that matters. The generative AI performance report answers whether your pages are being used in AI Overviews and AI Mode, which was genuinely unanswerable before, and it does so with the same dimensions as the conventional report. But it carries impressions alone, and Google's own documentation makes clear those impressions were already inside existing Search Console totals rather than hidden outside them. That makes this disaggregation rather than disclosure. Until clicks appear, exposure is measurable and arrival is not, and any figure purporting to quantify AI referral traffic is a model rather than a measurement.
Frequently Asked Questions
Does the report show clicks from AI Overviews?
No. As documented it carries impressions only, with no clicks, no click-through rate and no query data. Google has indicated more metrics may follow without committing to a timeline.
Were AI impressions previously hidden?
No. Google's help documentation states the report includes data from the Web search type in the Performance report, meaning these impressions were already counted inside existing totals. The report disaggregates them rather than revealing a new stream.
What counts as an impression here?
Google defines it as how many times links to your site were shown to a user in a generative AI feature on Google Search. It measures a link appearing, not a page being read or visited.
Can I compute a click-through rate for AI features?
Not defensibly. The clicks in your conventional report blend AI-driven and ordinary visits, so dividing those clicks by generative impressions produces a ratio without a coherent denominator.
How do I keep content out of AI Overviews?
Google documents nosnippet, data-nosnippet, max-snippet, noindex, robots.txt rules and the Google-Extended token. The snippet controls are blunt: suppressing a snippet also affects how the page presents in conventional results.
Are Search Labs experiments included?
No. Google states Search Console does not include data from experiments in Search Labs, so anything being tested with a subset of users does not appear in the report.
Sources
- official Google, Generative AI performance report (Search), Search Console Help, retrieved 26 September 2026. Source of the impression definition, the surfaces covered, the groupings, the row and preliminary-data limitations and the relationship to the Web search type.
- official Google, AI features and your website, Search Central documentation, last updated 10 December 2025. Source of the publisher controls and the statement that there are no additional technical requirements.
- trade press Trade coverage of the Search Central announcement of generative AI performance reports, consulted 26 September 2026 for the 3 June 2026 launch and the worldwide rollout by 31 August 2026. The announcement post itself returned only navigation on retrieval.
- official Google, Performance report (Search results), Search Console Help. Source of the four metrics and six dimensions in the conventional report. Read October 10, 2026.