Since June 3, 2026, Google has been rolling out a Search Console report that did not exist in this form before: the Generative AI report, available for all websites worldwide since August 31, 2026, according to Google. It shows how often your pages appear in Google's generative AI features: in AI Overviews and AI Mode, and, in a separate report, in generative Discover features. For the first time, Google delivers official numbers for a layer of visibility that was previously only measurable indirectly.

Before this report, AI Overview visibility could not be separated from classical impressions in Search Console: an appearance in an AI overview and a blue link counted into the same bucket, and in the overall Performance report they still do. Anyone who wanted to know whether their domain shows up in generative answers had to rely on their own prompt tests. That is changing now — partially.

Partially, because the report can do considerably less than the initial excitement suggests. And because a logging error in August showed how quickly a naive reading of the time series goes wrong. This article puts both into perspective: what is in it, what is missing, and how the report fits into a measurement routine that can carry decisions.

The timeline: from a UK test to a worldwide rollout

Google announced the report on June 3, 2026 on the Search Central blog and initially enabled it for a subset of websites only; according to Google, testing ran with website owners in the United Kingdom. Around August 11, the report became visible in many more properties, but according to John Mueller not yet in all of them. Google marked completion only with an added note: since August 31, 2026, the report has been rolled out to all websites worldwide. Anyone who has not actively checked may not have opened it to this day.

Jun 3

Announcement on the Search Central blog, rollout to a subset of properties

Aug 31

Rolled out to all websites worldwide, per a note added by Google

Aug 13–17

Logging error: impressions under-reported, data restored on Aug 21

The dates belong together: right during the broad expansion, a logging error under-reported impressions from August 13 to 17. Google restored the missing data on August 21. If you exported data during that window, you may still be working with wrong values. More on that below.

What does the Search Console Generative AI report show?

The Generative AI report shows exactly one metric: impressions, meaning how often links to your site appeared in generative AI features on Google Search. It covers AI Overviews and AI Mode; generative features in Discover have a separate report. Clicks, CTR, position and search queries are not included.

According to the Search Console Help Center, impressions can be broken down by four dimensions: pages (the canonical URL after redirects), countries, devices (desktop, tablet, mobile) and date, with hourly to monthly granularity per the announcement. In the chart, Google counts per property: if two links from the same site appear in one AI answer, that is one impression. The newest data is preliminary and shown as a dotted line. Data can be downloaded via the export button; Google has not documented API access so far. Since August 31, 2026, the report has been rolled out to all websites worldwide, but it only appears for properties with enough AI impressions.

This yields three first looks worth taking before any dashboard gets built:

1. Which pages carry the AI impressions?

The pages dimension answers whether AI visibility is spread broadly across the domain or hangs on a few URLs. Both have consequences: a broad distribution suggests that generative systems use the domain as a whole. A strong concentration shows which content types actually land in answers — and which investments went nowhere.

2. How do countries and devices break down?

If you operate internationally, you see here in which markets generative features already touch your visibility and where they do not. The device breakdown answers how AI impressions distribute across device categories — relevant to the question of which usage situations are affected.

3. How does the time series develop?

The time series is the genuinely valuable view — with hourly resolution for short-term effects and monthly resolution for the strategic line. It is also the view that has to be read most carefully: the August logging error showed that a kink in the curve can be a data-collection problem rather than a change in visibility.

What the report does not show — and what that means for interpretation

The gaps are bigger than the scope. An honest inventory:

Generative AI report · as of September 2026
Metric / featureIn the reportConsequence
ImpressionsYesThe only metric — shows appearance, not impact
ClicksNoNo traffic statement possible
CTRNoNo efficiency comparison with classical search
Queries / promptsNoNo intent mapping, no topic analysis
API accessNot documentedExport via button, no automated pipelines

Google has signaled additional metrics, without a timeline. Until they arrive, two hard limits of interpretation apply:

Without clicks, no traffic statement. An AI impression proves that a page appeared in a generative answer. Whether that appearance was noticed, read, or clicked cannot be derived from the report. Any conversion of AI impressions into visitors, leads, or revenue is, on this data basis, made up.

Without queries, no intent mapping. The report does not say for which questions the pages appeared. Whether a URL surfaces for purchase-ready prompts or for casual definition questions remains invisible — and with it the commercially decisive information.

From this follows the most important equation of this article: impressions ≠ citations ≠ visitors. Anyone presenting slides with “AI visibility +X%” on this basis is selling a number whose business meaning they do not know. Such reports are already circulating — they are the AI equivalent of the ranking report without search volume.

“The report does not answer the question of what AI search delivers for you. It answers the question of whether you are in the game at all — and on which pages.”

The August 13–17 logging error: what it teaches

For data from August 13 to 17, 2026, the report under-reported impressions. Google classified the error in its list of data anomalies as purely a logging problem: data collection was faulty, while actual visibility in AI Overviews and AI Mode did not change. On August 21, Google restored the missing data; the report now shows complete values for that period. There was also a one-day error on August 13 in the Discover report, including generative Discover features.

The operational risk of such errors lies less in the data than in its interpretation. Three wrong reactions are tempting: First, the panic analysis — a team sees the drop and launches a root-cause investigation for a problem that does not exist. Second, false causality — anyone who shipped a relaunch, a content migration, or a technical update around August 14 sees a kink that has nothing to do with it. Third, stale exports: if you saved data between August 13 and 21, your baseline still carries the faulty values. Pull those exports again. And while an anomaly is open, a real visibility loss cannot be told apart from the logging error either, in either direction.

Operator Insight

The annotation that will save you in six months

Every dashboard and every reporting document carrying these numbers needs an event annotation for August 13–17, 2026 (“GSC logging error, data restored by Google on Aug 21”). Search Console now offers its own annotations in the Performance report; in external dashboards you have to maintain them yourself, ideally with a regular look at Google's list of data anomalies. Time series without annotations generate discussions months later that nobody can resolve anymore, because by then nobody remembers the bug.

The practical routine: how the report belongs in your reporting

The report is not a dashboard replacement but a data source with narrow limits. Within those limits it is useful — if the routine is right. Four building blocks:

Step 1: Secure a baseline, manually, because no API is documented

Google documents no API access for the report; the Help Center only describes the export button for chart and table data. The table is subject to the same limits as the Performance report, such as 1,000 rows. So: define a fixed interval (weekly is enough for most organizations), file the export, document the export date and any known data anomalies. This sounds trivial and is the step most often skipped. Whoever starts exporting six months from now has no comparison base for the only question the report can answer: whether something is changing.

Step 2: Read trends, ignore absolute values

What exactly counts as one impression inside a generative interface cannot be audited from the outside — and without clicks and queries, the absolute value has no business translation. What is usable is the direction: is your own time series rising or falling, and is the structure shifting across pages, countries, and devices? Read your own history against itself, never against third-party benchmarks — for which no reliable data basis exists anyway.

Step 3: Hold the page level against your own citable assets

The analytically most productive exercise: lay the URLs with AI impressions against the list of pages deliberately built as citable assets — definitions, data, structured answers. Three patterns are possible. An asset collects impressions: it works. An asset collects none: a diagnostic case for structure, accessibility, or relevance. And a page collects impressions that was never meant as an asset: a hint at what generative systems actually use from the domain — often not what the content team expects. The systematic version of this comparison, including prompt tests against multiple models, is the core of the GEO audit.

Step 4: Combine with cross-model tracking

The report covers Google's ecosystem exclusively. For Microsoft Copilot, the AI Performance report in Bing Webmaster Tools has shown citations and grounding queries since February 2026. ChatGPT, Perplexity, and Claude still give site owners no comparable reports. For those systems, prompt-based tracking remains the only data source: repeated, versioned prompt sets against multiple models, evaluated for mention, portrayal, and competitors, as in LLM citation monitoring. For the qualitative diagnosis — why a model picks or passes over a source — the method from prompt reverse engineering complements this. The GSC report then validates a subset of these observations with official numbers; it replaces none of them.

The opt-out switch: a business decision, not an SEO trick

Alongside the report, Google introduced a Search Console toggle that lets sites take their content out of AI Overviews, AI Mode and AI Overviews in Discover. According to Google, this setting is not used as a ranking signal for search results outside generative AI features; excluded sites, however, no longer receive impressions or traffic from those features, so the report stays empty. The toggle has been available worldwide since August 31, 2026. For the first time, a separation exists that was missing until now: out of generative answers, in the classical index.

This decision is a business decision, not an SEO tactic. The switch is a candidate for business models whose value lies in the content itself — paid content, exclusive data, original research — and that do not want generative answers to preempt that value. Whoever lives on visibility, recommendation, and authority, by contrast, gives up with the opt-out exactly the surface this discipline is about. And the switch is not a ranking lever in either direction: it promises that organic rankings outside AI features remain unaffected; a mechanism by which an opt-out improves classical rankings does not exist.

Generative AI report FAQ

Why do I see no clicks in the Generative AI report?

Because the report does not include them. It reports impressions from generative AI features exclusively — no clicks, no CTR, no queries. Google has announced additional metrics without a timeline. Until they arrive, statements about traffic or efficiency based on this report are not possible, not even approximately.

Was the mid-August drop in AI impressions a warning sign?

No. For data from August 13 to 17, 2026, the report under-reported impressions due to a logging error. According to Google, this affected data collection only, not actual visibility. On August 21, Google restored the missing data. Pull exports from that period again and annotate it in your own reporting.

Does the report replace external AI monitoring?

No. It covers Google's ecosystem exclusively: AI Overviews, AI Mode, and generative Discover features. Microsoft Copilot has had an AI Performance report in Bing Webmaster Tools since February 2026; for ChatGPT, Perplexity, and Claude there are no comparable reports for site owners. There, prompt-based cross-model tracking remains the only measurement method. The report adds official Google numbers to that tracking; it cannot replace it.

Bottom line

The Generative AI report is real progress: for the first time, Google separates AI visibility from classical impressions and makes it visible per page, country, device, and time range. It is at the same time limited progress: a single metric, no documented API access, no queries. And as the August logging error showed, the time series is easy to misread without an eye on Google's data anomalies.

Used correctly, it is the Google module of a larger measurement architecture: a manually secured baseline, reading trends instead of absolute values, comparison against your own citable assets, combined with cross-model tracking for the systems without analytics. Used incorrectly, it is a percentage generator for slides nobody can stand behind.

Set up the routine now, and in twelve months you will have an annotated, interpretable time series. Everyone else will have screenshots.

Sources