What can Google Search Console measure, and what remains unknown?

Start with the boundary of each dataset

Search Console can answer questions about URL status and performance in Google Search. On its own, it cannot prove why an AI answer mentioned a brand, what result a content change caused, or what every user saw. The data surface, date range, aggregation method and unknown states must remain separate.

For a team establishing a measurement system, the first task is not to find an “AI score”. It is to understand what each dataset can and cannot answer. This prevents an unprocessed URL, a temporarily small sample, an empty interface state or a third-party estimate from being presented as Google’s internal judgement.

Four data surfaces and the questions they answer

This table is wide. On a touch device, swipe left or right to view every column.

Google’s official guidance on AI features and websites explains that AI Overviews and AI Mode still depend on normal Google Search indexing, snippet eligibility, quality and technical requirements. Eligibility does not guarantee crawling, indexing, serving or citation.

If enabled, Cloudflare Web Analytics is a separate surface for aggregate visits and performance. It is not part of Search Console and does not report AI answers or Google Search generative AI results. Keep it separate rather than adding it to the table or a combined “AI result”.

The substantive change on 31 August 2026: global rollout

Google launched the Generative AI performance reports on 3 June 2026. Its announcement and Search Console Help page now state that, as of 31 August 2026, these insights have rolled out to all websites worldwide. The previous wording that Google was gradually providing the report to only some properties is therefore no longer this article’s current availability baseline.

“Rolled out worldwide” is Google’s statement about product rollout; it is not a guarantee that every account must display the report or that any named property has reportable volume. The “Not seeing the report?” section on the same Search Console Help page still contains transitional wording that “not all properties have access” because the report is “rolling out over time”; it also lists insufficient impressions or exclusion from Search generative AI features as possible reasons. If an account still does not show the report, verify the correct property, account state, feature eligibility, impression volume, date range and filters, and preserve this unresolved wording in Google’s own page. The global-rollout note alone cannot establish that CLEARCRAFT or a client site has reportable impressions.

What the Generative AI performance report includes

The Generative AI performance report (Search) currently includes impressions from AI Overviews and AI Mode on Google Search. The report can be read through these dimensions:

• Pages: grouped by the final URL linked from a generative AI feature; data is generally assigned to the canonical URL.
• Countries: grouped by the country where the search originated.
• Dates: viewed by day, week or month; all dates use Pacific Time (PT).
• Devices: grouped as desktop, tablet or mobile.

Search Console does not include data from experiments in Search Labs. The report provides impressions as officially defined by Google; impressions do not reveal the answer text, citation rationale, ranking position, clicks, enquiries, revenue or causation from a content change.

Limits when reading charts, tables and exports

The newest data may be marked preliminary, meaning that collection is still in progress and the values can change. Retain this status when reading the latest dates instead of comparing an unstable value directly with a completed period.

The chart total is aggregated by property; after a URL filter is applied, chart data is aggregated by URL. Table aggregation depends on the selected dimension: country, device and date are aggregated by property, while page data is aggregated by page. Chart and table totals can therefore differ because of property-versus-page aggregation. Do not interpret the difference as missing or erroneous data without further evidence.

The usual Search performance report limitations also apply, including the typical 1,000-row limit, available time periods and other reporting limits. When data is downloaded, values shown as ~ or -—not available or not a number—in the interface become 0 in the export. Preserve the original interface state or a limitation flag so that a format-generated zero is not mistaken for a true zero-impression result.

The report’s data is already included in the Web search type of the Performance report (Search results). Read general Web performance and the dedicated generative AI view separately, but do not add them together or treat them as non-overlapping user groups.

A visible report with no data does not mean zero impressions or a nonexistent feature

A visible report entry without a value can reflect insufficient volume, the selected date range or filters, preliminary processing, the site’s inclusion state, or display and export limitations. A blank, ~, - or exported 0 is not enough to prove a true zero-impression result. It also does not prove that AI Overviews, AI Mode or the report itself does not exist.

Internal records should retain the observed state—such as not available, insufficient volume, preliminary, filtered or export limitation—instead of defaulting everything to zero. Record zero only when the Google interface supports that value for the verified property, date range, filters and data state. Even then, zero impressions do not establish that the site never appeared at another time, in another country or device category, or in another AI product.

Two data-quality boundaries that still matter

URL Inspection shows information about Google’s indexed version by default; Live Test checks whether the current URL can be accessed and processed. Both are worth recording, but a positive Live Test does not mean that the page has appeared in Google Search.

Google recorded a logging error in the Search Generative AI performance report from 13 to 17 August 2026. In an update dated 21 August, Google states that the missing data for that period was restored and that the Performance report now shows complete metrics. This is a resolved historical anomaly: the period should no longer be described as currently incomplete or automatically labelled an ongoing data_quality_exception. The method that remains useful is to check Google’s official Data anomalies page and restoration notes before interpreting a sudden rise or fall; the chart shape alone cannot attribute a change to content or technical work.

A third-party GEO score can support workflow, but it is not Google internal ranking data

Google states that third-party SEO, AEO and GEO tools do not have access to Google’s internal ranking data and cannot guarantee ranking success or performance. A third-party GEO score may support diagnosis, work prioritisation or retesting within the tool’s own methodology, but it must retain the provider, method, date, sample, limitations and human-review state.

It must not be renamed a “Google AI score”, “Google internal score” or Search Console metric, and it cannot fill gaps in Search Console data. A tool’s prediction remains the tool’s prediction. Formal interpretation should prioritise current official Google documentation, first-party Search Console data and preserved raw observations.

Conclusions that the data cannot support, and the next step

Search Console cannot, on its own, prove why an AI answer mentioned a brand, what result a redesign caused, or what every user saw. Google’s public guidance for generative Search remains grounded in normal technical foundations and useful, non-commodity content for people. No AI-specific file, special markup, single GEO score or one tactic can guarantee inclusion, citation, traffic or revenue from generative features.

FAQ, schema and sitemaps should therefore serve their normal purposes in content understanding, information quality and indexing. They should not be sold as Google AI shortcuts. A team can next review CLEARCRAFT’s evidence-led methodology, the evidence-first measurement note and service deliverables and data boundaries, build a reviewable recurring record, and then decide which content or technical hypothesis deserves a test.