Measurement brief · Search Console
Four boundaries to set before reading the chart
The new report is useful only when its impression is treated as a precisely defined exposure event—not as shorthand for citations, clicks or AI-search success.
- It measures linked visibility
An impression is recorded when a link to your property appears in an AI Overview or AI Mode response. The report does not reveal every sentence, source-selection step or unlinked mention.
- It is an impression-only lens
Dates, countries, devices and canonical pages are available. Queries, clicks, CTR and position are not.
- Chart and page totals answer different questions
Two links from the same property can create one property impression in the chart and one impression for each canonical URL in the page table.
- An absent report is not proof of zero exposure
Access, rollout, minimum-impression thresholds and the site’s generative-AI control must be checked first.
Google began rolling out a dedicated Generative AI performance report in Search Console in June 2026. It separates impressions from AI Overviews and AI Mode from the conventional Web results report and makes that exposure explorable by page, country, device and date.
The new view closes one measurement gap, but it does not turn Search Console into a citation tracker. The most important editorial and analytical decision is therefore not which line moved. It is what kind of event the line represents.
What Search Console can observe
A Google-served link to a canonical page, counted as an impression and grouped across time, country and device. This is a distribution signal: your page entered a visible response surface.
What remains outside
The prompt, wording around the link, quoted passage, competitor context, click, downstream session and commercial outcome all require other evidence.
The report is a visibility sensor, not a citation ledger
Google’s documentation defines the report around impressions from two production experiences: AI Overviews and AI Mode. Search Labs experiments are excluded. That distinction matters because third-party tools often use “AI visibility” to combine several different events: a brand mention, a quoted passage, a linked citation or simply a URL detected in a generated answer.
Search Console reports the event Google can associate reliably with your verified property: a link impression. This makes the data first-party and operationally valuable, but narrower than the label “citation” suggests. A rise can support the statement that linked exposure increased. It cannot, by itself, support the statement that Google quoted the site more often, sent more visits or preferred it over a competitor.
This is the same discipline needed when working with conventional ranking data. As our analysis of rank tracking without a public Google ranking API explains, a metric becomes misleading when the collection surface is allowed to stand in for the entire search experience.
One exposure can legitimately produce two totals
The property-level chart and the page table use different aggregation rules. When two links from the same property appear in one generative feature, the property is counted once in the chart. In the page table, each unique canonical page receives its own impression. Adding page rows can therefore produce a number larger than the chart total.
The chart answers: was this verified property represented in the exposure?
The table answers: which unique canonical URLs were represented?
This is not data loss or duplication. It is a change of unit. The correct response is to label the unit in every dashboard and comparison. If an executive view uses the property chart, its denominator is property exposure. If an editorial review uses page rows, it is analysing the distribution of canonical URLs within those exposures.
Canonicalisation also matters. Search Console assigns page reporting to Google’s selected canonical URL. Teams debugging unexpected rows should inspect canonical selection before assuming that the generative system preferred a different article. This connects to the broader search-infrastructure problem described in our guide to Googlebot link extraction and rendering failures.
A claim-boundary decoder for every change in the chart
The report becomes useful when every observation is paired with the narrowest defensible claim and an explicit evidence gap. This is the article’s operating module: read from the measured event toward the claim boundary, then decide which outside dataset must close the gap.
AI visibility claim boundary
Start with what changed. Stop before the report’s blind spot. Add the next dataset only when the decision requires it.
You may say linked exposure increased. Do not claim clicks, traffic, quoted text or citation share rose without corroborating data.
You may say different canonical URLs gained distribution. Do not claim a topic or query caused the shift because query data is unavailable.
You may say exposure distribution changed by that dimension. Do not infer user intent, conversion quality or market demand from the report alone.
You may say the reporting condition is unresolved. Do not conclude that generative visibility is zero until access, threshold and controls are checked.
For traffic impact, connect Search Console dates to web analytics and server logs. For engagement, examine landing-page sessions and downstream events. For commercial value, use qualified leads, revenue or another outcome outside Search Console. For source wording and competitor context, a separate observation system is required—and its collection limitations must remain visible.
That evidence stack also protects against a familiar content strategy error: assuming that more exposure proves more information gain. Our analysis of information gain in AI-mediated search explains why distinctiveness is a property of the material supplied, not a conclusion that can be read directly from an impression line.
Why the report may be missing
Google lists three practical reasons a property may not show the report. The rollout may not yet include the account, the property may not have enough impressions to meet reporting thresholds, or the site may have excluded its content from generative AI features. These possibilities require different responses and should not be collapsed into a zero.
Confirm the property owner can see the report and that it has reached this Search Console account.
Insufficient reportable impressions can suppress the view even when isolated exposures exist.
An exclude directive can remove the site from AI features after Google processes the live control.
The include state is the default. Google says an exclude control affects links and grounding in AI Overviews and AI Mode but does not change ranking in other Google Search experiences. Changes generally take one or two days after the control is live, although some can take longer. That delay makes same-day absence a weak diagnostic signal.
Three details that prevent false trend stories
These are small interface details with large analytical consequences. A day-over-day anomaly can be a timezone boundary. A recent drop can be incomplete processing. A zero in an export can be a transformed display value rather than a measured absence. The safe operating rule is to document extraction time, timezone and whether the final data point was still preliminary.
Search Console also includes generative AI data inside the overall Web Search performance report. Analysts comparing the dedicated report with older Web totals must therefore avoid treating the two as independent pools and adding them together.
A compact weekly decision protocol
Use the dedicated report to detect changes in exposure, not to manufacture certainty. First, compare stable date windows and exclude preliminary days. Second, identify whether the movement comes from pages, countries or devices. Third, inspect canonicalisation and known technical changes. Fourth, join the dates to analytics or server evidence. Finally, phrase the conclusion at the level the combined evidence supports.
If the operational question is technical health, keep that investigation separate from the AI report. Search Console can be a useful diagnostic surface, but its limits are described in our examination of Search Console as a vulnerability detector. A clean generative AI chart is not proof that crawling, rendering, security or measurement infrastructure is healthy.
The new report is valuable precisely because it is narrow. It creates a first-party baseline for one previously hidden exposure surface. Used with a disciplined claim boundary, it can show where Google’s generative experiences are distributing a site’s links—and where a team must resist filling the remaining darkness with assumptions.
Frequently asked questions
Does the report show AI Overview and AI Mode clicks?
No. The dedicated report currently provides impression data. Clicks, CTR and position are not available in this view.
Why do page impressions exceed the chart total?
The chart counts the property once per exposure, while the page table counts each unique canonical URL represented. Two linked pages can therefore produce one property impression and two page impressions.
Does excluding content from generative AI lower normal Search rankings?
Google says the control affects links and grounding in generative AI features, not ranking in other Google Search experiences. Processing generally begins within one or two days, with some cases taking longer.
Sources
Primary Google documentation checked 13 August 2026
- Google Search Central Blog: Generative AI performance reports in Search Console
- Search Console Help: Generative AI performance report
- Google Search Central: AI features and your website
- Google Search Help: Control content in Search generative AI experiences
- Search Console Help: Performance report data and aggregation
- Search Console Help: Troubleshooting performance data discrepancies
