Search Console has separate limited-rollout reports for impressions from generative AI features in Google Search and Google Discover. The Search report currently covers AI Overviews and AI Mode. The Discover report covers supported generative AI features inside Discover.
Both dedicated reports are impression-only. They do not provide AI-specific clicks, CTR, queries, prompts, position, generated response text, citation order, or conversion data.
This guide reflects Google's documentation on July 18, 2026.
The current reporting map
| Report | Included context | Metrics | Dimensions |
|---|---|---|---|
| Generative AI Performance — Search | AI Overviews and AI Mode | Impressions | Page, country, date, device |
| Generative AI Performance — Discover | Generative AI features in Discover | Impressions | Page, country, date |
Google says the included feature list can change as Search develops. Search Labs experiments are excluded from both reports while they remain in active development.
Do not combine Search and Discover impressions without retaining the surface label. Their aggregation and impression-counting rules differ.
Availability is deliberately limited
Google is rolling the reports out to a subset of website owners for testing before broader availability.
A report can be missing because:
- The property is not included in the rollout.
- The site has not reached the unpublished minimum impression threshold for that generative AI surface.
- The selected Search Console property does not contain the credited canonical URLs.
- The site is excluded through the Search generative AI control.
- The selected date range predates reportable evidence.
- Supported features have not shown links to the site.
The numeric threshold is not documented. A missing report does not prove zero generative AI visibility, and seeing one of the two reports does not guarantee access to the other.
Search generative AI report
The Search report includes impressions for:
- AI Overviews.
- AI Mode.
It excludes Search Labs experiments and can add other production features in the future.
Google states that the report's data is included in the Web search type of the standard Search results Performance report. Therefore the AI report is a dedicated view into part of Web Search impression activity, not an extra source to add on top of Web totals.
Search impression
An impression records that a link to the property was shown to a user in a supported generative AI feature on Google Search.
The dedicated report does not show:
- Whether the link was clicked.
- Which query or prompt produced it.
- Where the link appeared inside the generated response.
- Whether the page's text was quoted or paraphrased.
- Whether the site was used only for grounding without a visible link.
- How many other sources appeared in the response.
- A ranking or citation position.
Call the metric linked-page impressions, not “AI citations,” “mentions,” “answers,” or “visits.”
Search report aggregation
Chart
The chart is property-aggregated. If two results from the same property appear in one generative AI Search feature, they count as one property impression in the chart total.
Adding a URL filter changes the chart to URL aggregation.
Table
Table aggregation depends on the selected dimension:
- Country, device, and date are aggregated by property.
- Page is aggregated by page.
Consequently, summing page rows can differ from the property-level chart. Several linked pages from one property can each receive a page impression while the chart counts one property impression for the feature.
The usual Search Performance limitations, including the 1,000-row interface table limit, apply.
Search report dimensions
Page
Page groups data by the final URL linked from the feature after redirects. As with most Performance data, Google generally assigns the metric to the selected canonical URL rather than a duplicate.
This can shift evidence:
- Between HTTP and HTTPS properties.
- Between mobile and desktop variants.
- Across subdomains.
- From a syndicated URL to a cross-domain canonical.
- Away from the URL that a browser ultimately displayed.
Use URL Inspection to confirm canonical attribution before declaring a page absent from the report.
Country
Country describes where the Search originated. It does not reveal language, prompt wording, residence, or market intent.
Device
Device is desktop, tablet, or mobile. It does not identify browser, operating system, AI feature subtype, or signed-in status.
Date
Date supports daily, weekly, or monthly grouping and uses Pacific Time. Newest data can be preliminary and is marked with a dotted line.
Discover generative AI report
The Discover report measures impressions for links shown in supported generative AI features inside Discover.
It is separate from:
- The standard Discover Performance report.
- The Search generative AI report.
- Search Labs experiments.
The dedicated Discover AI report provides page, country, and date dimensions. It does not currently document a device dimension.
Discover AI impression rules
A Discover generative AI link must be scrolled into view, either as a standard card or within a carousel.
Only one impression is counted for the same result during a session if the user scrolls away and back.
Discover AI data is page-aggregated. If two different results from the same property appear in one Discover list, each can receive its own impression.
This differs from the Search AI chart's property aggregation. Do not compare the two totals as though they use the same counting unit.
Discover report dimensions
Page
The page is the canonical URL that served as the source shown to the user, not necessarily the URL where the user lands after clicking. If the canonical belongs to another Search Console property, the alternate URL is absent from the current property's table.
Country
Country is where the content was viewed. It does not expose interest, language, profile, or prompt data.
Date
Dates follow the standard Discover report's time-zone and granularity rules. Daily views use Pacific Time; the newest data can be preliminary.
Thresholds and table limits
The Discover AI chart totals are accurate for the selected period and are not truncated. Individual daily points can be omitted when they do not reach a reporting threshold, while remaining included in the total.
The table inherits the standard Discover report's 1,000-row interface limit.
Therefore:
- Visible page rows may not sum to the chart total.
- A low-volume day can contribute without appearing as a plotted point.
- Reversing table sort does not reveal row 1,001.
- A UI export does not turn the table into a complete URL inventory.
Why there are no AI-specific clicks
The dedicated reports currently expose impressions only. They do not provide an AI-specific click metric or CTR.
The Search AI documentation says its data is included in the standard Web search type. This means Web Performance remains the broader source for Search clicks and impressions, but it does not supply a documented filter that isolates clicks caused specifically by AI Overviews or AI Mode.
You therefore cannot calculate:
- AI Overview CTR.
- AI Mode CTR.
- Clicks per AI-linked page impression.
- Conversion rate from AI features.
- Click share by AI feature.
Do not divide standard Web clicks by dedicated AI impressions. The numerator and denominator describe different populations.
Why there are no queries or prompts
Neither report provides a query, prompt, follow-up question, or topic dimension.
The Search report can show which pages received linked impressions, but it cannot answer:
- What the user asked.
- Whether an AI Overview or AI Mode produced the impression when both are included.
- Which passage or entity matched the request.
- Whether the request was branded or non-branded.
- Which prompt variation caused a change.
Standard Web query data cannot be joined deterministically to an AI impression. Matching by page, date, country, and device can support a hypothesis, but it does not reconstruct the missing query-to-AI event relationship.
Why “citation” is the wrong unit
Google documents the metric as impressions for links to the site shown in a generative AI feature.
The report does not expose:
- Citation count.
- Citation position.
- Link anchor or surrounding text.
- Generated answer text.
- Source carousel order.
- Whether the page was paraphrased.
- Grounding without a visible link.
Use “generative AI linked impression” in reporting. If another tool calls an observed link a citation, keep that third-party observation separate from Search Console's metric.
Search generative AI control
Search Console is also rolling out a property setting under Settings → Search generative AI.
It manages eligibility for:
- AI Overviews.
- AI Mode.
- Generative AI features in Discover.
The available choices are:
| Control | Effect |
|---|---|
| Include | Links and content can appear and help ground supported Search generative AI responses. This is the default for all properties. |
| Exclude | Links and content are prevented from appearing or grounding responses in the covered Search generative AI features. |
| Inherit control from parent | The property follows its nearest configured parent property. This is the default for a child property. |
An owner can override inheritance for a child property.
Property inheritance
The closest parent that has explicitly stopped inheriting supplies the control value. A parent can be:
- A top-level domain property.
- A subdomain property.
- A broader URL-prefix property.
A path property can inherit from a broader URL-prefix or domain property and can then override the inherited choice for its own path and descendants.
Audit parent and child settings before changing the control. An exclusion set high in the hierarchy can affect many properties.
What exclusion does and does not do
Excluding a property prevents its links and content from appearing to users or grounding responses in the covered Search generative AI features.
Google states that the control:
- Is not a ranking or inclusion signal for other parts of Search.
- Does not override participation in Merchant Center or Google Ads.
- Does not block all use of the site in Search systems, such as understanding language.
- Does not control AI model training.
To remove a page from Google Search completely, use an appropriate noindex method. To manage qualifying model-training and Gemini/Vertex grounding uses, Google documents the separate Google-Extended robots token.
These controls are not interchangeable.
Search control versus Google-Extended
| Control | Primary scope |
|---|---|
| Search generative AI setting | Visible links, content, and grounding in covered AI features of Google Search and Discover |
Google-Extended |
Training future Gemini models plus grounding in Gemini Apps and Vertex AI through Google Search |
noindex |
Page eligibility for Google Search generally |
| Googlebot crawling rules | Whether Google can crawl content for Search indexing and Search features |
Google-Extended has no separate HTTP user agent; it is a robots.txt token used as a control. Google says it does not affect Google Search inclusion or serve as a Search ranking signal.
Consult legal, product, SEO, and data-governance owners before changing any of these controls.
Timing after a control change
Google says a Search generative AI setting change generally takes a few days to propagate. Once the new control is live, exclusion normally happens within one to two days, although cached content or system propagation can take longer.
Do not judge the effect from the same day's chart. Record:
- Property and inherited setting.
- Change timestamp.
- First day the control is reported live.
- Expected propagation window.
- Preliminary-data boundary.
- Search and Discover AI trends separately.
An impression decline after exclusion confirms less reported visibility, not the amount of traffic or conversions lost because the dedicated reports do not provide clicks.
Report export and automation boundaries
Both dedicated reports provide a UI export for their chart and table. Values shown as ~ or - become zero in exported data.
Google's current public Search Analytics API documents web, image, video, news, discover, and googleNews types. It does not document a generative-AI type, feature dimension, or appearance filter.
The current BigQuery bulk-export schema likewise has no dedicated generative-AI search type or appearance field.
Therefore automation cannot currently retrieve an official isolated AI-impression series through the public API or bulk export. A scheduled browser scrape is not a stable or recommended substitute for an API.
Automation can still:
- Store periodic manual UI exports.
- Track the broader Web and Discover performance series.
- Reconcile AI-linked pages with first-party content metadata.
- Monitor canonical and index status.
- Annotate control changes and Google updates.
- Calculate concentration, page cohorts, countries, and device mix from exported report rows.
Keep official AI impressions separate from third-party SERP monitoring and model-prompt experiments.
A reliable analysis workflow
1. Confirm report and property eligibility
Record rollout access, selected property, parent control, canonical scope, and the relevant Search or Discover report.
2. Preserve the native total
Capture the chart total before segmenting. For Search, note whether it is property-aggregated or URL-filtered. For Discover, note that it is page-aggregated.
3. Segment only supported dimensions
Search:
- Page.
- Country.
- Device.
- Date.
Discover:
- Page.
- Country.
- Date.
Do not invent feature, prompt, query, or citation dimensions.
4. Compare matched periods
Use finalized Pacific-Time periods, align full weeks or months, and annotate report rollout or control changes.
5. Analyze concentration
Measure:
- Share of impressions from the top page.
- Share from the top ten pages.
- Number of pages with visible report rows.
- Country and device concentration.
- New, growing, declining, and disappearing page cohorts.
Remember the 1,000-row and threshold limits.
6. Add broader context without claiming causality
Compare AI impressions with:
- Web or Discover impressions.
- Page indexing and canonical changes.
- Content releases.
- Country and device shifts.
- First-party landing-page sessions and conversions.
These comparisons can guide investigation but cannot identify AI clicks or prompts.
Interpreting common patterns
AI impressions rise while Web impressions remain stable
Possible explanations include a larger share of existing Web exposure occurring in supported AI features, feature rollout, or changed feature selection. The reports do not prove which explanation applies.
Do not add the AI increase to Web impressions; the Search AI series is included in Web data.
AI impressions fall while Web clicks remain stable
The site may have lost AI-linked exposure while other Web results offset it, or aggregation and feature mix may have changed. The dedicated report cannot show whether AI clicks changed.
One page dominates AI impressions
Treat the property trend as concentrated. Separate that page before claiming site-wide generative visibility growth.
Search AI appears but Discover AI does not
The reports have separate rollout, surface eligibility, and thresholds. This is not inherently an error.
The report disappears after exclusion
Confirm the parent/child control, propagation window, date range, and whether previous historical data remains visible. Do not treat absence as a measurement of clicks lost.
Common mistakes
Calling impressions “AI traffic”
An impression is a shown link, not a click, session, or user.
Reporting citations
Search Console does not expose citations, quotation use, response text, or source order.
Calculating AI CTR from Web clicks
The metrics do not share an isolated event population.
Adding AI Search impressions to Web totals
Google says the Search AI data is included in the Web search type.
Combining Search and Discover AI totals without labels
They use different surfaces, dimensions, and aggregation.
Treating a missing report as zero visibility
Rollout and unpublished thresholds can hide the report.
Blocking Google-Extended to opt out of Search AI
Google-Extended does not affect Search inclusion. Use the Search generative AI property control for the covered Search features.
Assuming the control governs AI training
Google explicitly says it does not. Training controls are separate.
What Search Console can and cannot conclude
It can show:
- Linked-page impressions from supported Search generative AI features.
- Linked-page impressions from supported Discover generative AI features.
- Page, country, date, and—on Search—device breakdowns.
- Canonical attribution.
- Property-versus-page aggregation behavior.
- Whether the property has access to the reports and control.
It cannot show:
- AI-specific clicks, CTR, or conversions.
- Queries, prompts, or follow-up questions.
- AI Overview versus AI Mode as separate rows.
- Generated response text.
- Citation count, order, anchor, or context.
- Grounding without a visible link.
- Search Labs experiment data.
- A public API or BigQuery AI dimension.
- Why a page was selected.
- Whether an impression change caused a traffic change.
Review checklist
- Search and Discover generative AI reports are kept separate.
- Search is understood to include AI Overviews and AI Mode.
- Search Labs experiments are excluded.
- The metric is called impressions, not traffic or citations.
- No AI-specific clicks, CTR, queries, prompts, or position are inferred.
- Search chart property aggregation is distinguished from page rows.
- Discover page aggregation and scroll-into-view rules are documented.
- Canonical-property attribution is verified.
- The 1,000-row and threshold limits are considered.
- Search AI impressions are not added to Web totals.
- Report rollout and minimum-impression eligibility are considered.
- Parent and child control inheritance is audited.
- Search control, Google-Extended, Googlebot, and
noindexare not conflated. - UI exports are distinguished from API and BigQuery availability.
- Third-party AI-monitoring estimates remain labeled separately.
Related content
- The Performance report: complete guide
- Discover Performance: complete guide
- Search Console Insights: complete guide
- Search Console API: complete guide
- BigQuery bulk export: complete guide