Which Google features are included in the report?

Google's help page currently lists AI Overviews and AI Mode. Search Labs experiments are excluded, and Google says the supported-feature list may change.

Does the report show the prompts or queries that triggered an AI result?

Google's documented dimensions are page, country, date and device. Query or prompt is not listed, so teams should not infer prompt-level performance from the report.

Is an impression the same as a visit or citation?

No. An impression means a link to your site was shown to a user in a supported Google generative AI feature. Visits require analytics data, while citation and mention tracking use different definitions and collection methods.

Do third-party AEO tools become unnecessary?

No, but their role is narrower and clearer. Search Console owns first-party Google impression data; third-party tools can sample named prompts across engines, preserve answer context and support diagnosis. The two datasets should remain separately labelled.

Google Search Console's Generative AI Report: What It Measures and What It Misses

On 31 August 2026, Google completed the worldwide rollout of dedicated generative AI performance reporting in Search Console. Site owners now have a first-party view of links appearing in AI Overviews and AI Mode instead of trying to reconstruct that visibility from referral strings, rank trackers or screenshots.

That is a substantial improvement. It is not a complete AEO dashboard.

The distinction matters because first-party and sampled data answer different questions. Search Console can tell you that Google showed a link. It does not publish the full answer context or provide a cross-engine comparison. Treating either system as a substitute for the other produces confident charts with confused definitions.

Disclosure: AEO Landscape and CiteCue have common ownership. We discuss CiteCue as one option for the cross-engine layer of this workflow. It has no access to Google's internal ranking data, and its observations must not be represented as Search Console data. See our full disclosures.

What the report measures

Google's help documentation defines an impression as a link from your site being shown to a user in a supported generative AI feature on Google Search. The current supported list contains AI Overviews and AI Mode. Search Labs experiments are excluded.

The report provides five useful views:

Total generative AI impressions for the property over time
Pages whose canonical URLs received those impressions
Countries where the searches originated
Devices used for Search impressions
Daily, weekly and monthly trends, with dates reported in Pacific Time

The announcement also described hourly granularity. The current help page documents days, weeks and months in the date dimension, so the help documentation should be treated as the operating reference until Google resolves that difference.

There is another subtlety in the totals. The chart aggregates by property: if two results from the same site appear in one generative AI search feature, they count as a single property impression in the chart. The page table aggregates by page. Chart and table totals can therefore differ without either being wrong.

The usual Search Console limits also apply, including the 1,000-row table limit and preliminary newest data that may still change.

What Google does not document in the report

The published dimensions do not include:

The user's query or the model's fan-out queries
The generated answer text around a link
Brand mentions that contain no link
Competitors appearing in the same answer
A citation's order or prominence
Sentiment or factual accuracy
Results from ChatGPT, Gemini outside Google Search, Perplexity or Claude

This is not a criticism of the report. It is a boundary. Google is reporting its own impression data, not recreating the workflow of a prompt-monitoring platform.

It also means several tempting inferences are invalid. A rising impression curve does not identify which prompt family improved. A page with fewer impressions is not necessarily a worse page; it may address a smaller demand set. An impression is not a visit, a conversion, a brand mention, or proof that the generated answer endorsed the page.

Use three measurement layers

A defensible AEO dashboard keeps three systems separate.

Layer 1: Search Console for Google visibility

Use the generative AI report to answer: Did Google show links from our site, which pages appeared, where, and when?

Export the data on a fixed monthly date. Record the selected filters, property type, timezone and any known reporting gaps. Compare like with like rather than mixing partial weeks or changing country filters.

Layer 2: prompt sampling for answer context

Use a repeatable prompt set to answer: Were we named or cited for the buying questions we care about, what did the answer say, and who appeared instead?

This can be done manually at small scale. Once the set spans several engines and requires repeated samples, a monitoring platform becomes an operational convenience rather than a source of privileged engine data.

With CiteCue, the appropriate workflow is to keep a fixed prompt set, monitor cross-engine mentions and citations, inspect the factors attached to recurring losses, send supported changes into the fix queue, and re-run the same measurement after publication. Do not paste the resulting citation rate into a chart labelled "Google AI impressions." They are different denominators collected by different systems.

Layer 3: analytics for business outcomes

Use web analytics and conversion data to answer: Did people visit, engage and complete a useful action?

An AI impression can create value without a click, and a click can fail to create value. That is why impression, visit and conversion belong on adjacent lines rather than being compressed into a single visibility score.

A monthly operating routine

The following sequence is enough for most small teams:

Export Search Console generative AI impressions by page for the completed month.
Mark the pages with meaningful gains, losses and new visibility.
Compare those movements with the fixed prompt sample, keeping Google and non-Google engines separated.
Read the actual answer context for the largest disagreements.
Select one supported intervention: fix access, correct a fact, strengthen a weak passage, add original evidence, or improve a relevant third-party source.
Record the changed URLs and publication date.
Recheck the same measures after enough time for crawling and sampling.

The value is in the join between systems, not in pretending they are the same system. Search Console supplies authoritative Google evidence. Prompt monitoring supplies controlled observations. Analytics supplies outcomes. A change log ties all three to work your team actually shipped.

How to read the first three months

Do not set a target before observing the reporting baseline. The worldwide report is new, and sites differ radically in demand, eligibility and page mix. Three complete months will tell you more than a generic benchmark.

Watch for durable patterns rather than daily noise:

A small set of pages accumulating most impressions
New pages entering the report after a documented change
Country or device differences large enough to affect content decisions
Search Console visibility rising while prompt-level citations remain flat
Prompt-level citations rising on non-Google engines while Google impressions remain flat

The last two are not contradictions. They are evidence that the surfaces and collection methods differ.

Google has finally given AEO teams a reliable first-party line for one important surface. The mature response is to use it aggressively, label everything else honestly, and stop asking one metric to explain an entire answer ecosystem.

Sources and verification

Sources checked 14 September 2026. This article was researched, drafted and published through AEO Landscape's automated editorial workflow. The factual baseline comes from the primary sources above; interpretations and the three-layer model are ours.