Fact-checked 17 July 2026: Google Search Console now has dedicated generative AI performance reports for Google Search and Discover. They can show impressions by page, country, device, and date, but they do not currently expose clicks, click-through rate, or queries inside those reports. The data is useful for measuring where AI-result visibility is appearing. It is not a complete AI traffic, conversion, or citation-attribution system.
Short answer: use the report to build page cohorts and monitor AI-result exposure over time. Then use conventional Search Console data to diagnose queries, GA4 to evaluate landing-page behaviour, and a controlled manual prompt log to observe answers and citations. Do not claim that generative AI impressions caused a lead or sale unless you have additional evidence.
What is the Search Console generative AI performance report?
The generative AI performance report is a Search Console view that isolates impressions associated with Google’s generative AI experiences. Google announced separate reporting for Search and Discover. In each surface, the report can break impressions down by:
- page;
- country;
- device;
- date.
Google’s official announcement says the same impressions also remain part of the site’s overall performance data, so the dedicated report is a diagnostic slice rather than a separate universe of impressions. Google’s Search Console help documentation controls the current definitions and limitations.
Google describes a staged rollout to a subset of sites. Search Labs activity is not included. A missing report or zero values therefore does not automatically mean that nobody encountered the site in an AI experience.
What the report includes – and what it does not
| Available in the dedicated report | Not currently available there |
|---|---|
| Generative AI impressions | Clicks |
| Pages associated with those impressions | Click-through rate |
| Country breakdown | User queries or prompts |
| Device breakdown | The exact generated answer |
| Date trend | Whether the page was linked, cited, or merely contributed to an impression under Google’s definition |
| Separate Search and Discover views | Lead, transaction, or revenue attribution |
This boundary is the most important part of the report. An impression is evidence of exposure under Google’s reporting rules. It is not evidence of a visit, an engaged reader, brand recall, a conversion, or incremental revenue. Treating those outcomes as interchangeable creates measurement debt.
How Google counts generative AI impressions
Use Google’s definition rather than inventing a proxy. The reporting system decides when a site receives an impression in a generative AI experience, and the dedicated view aggregates those impressions across the dimensions above. The report does not reveal the query that triggered the experience or the sentence in which a source appeared.
This creates a deliberate evidence gap:
- You can observe that a page accumulated generative AI impressions.
- You can observe the time, country, and device mix.
- You cannot use this report alone to know the triggering query or prompt.
- You cannot use this report alone to know how the answer represented your brand.
- You cannot use this report alone to attribute a downstream conversion.
A good analysis preserves that chain instead of filling the missing steps with assumptions.
Search and Discover should be analysed separately
Search reflects active information seeking. Discover is a personalised feed where the user may not have expressed a query at that moment. The same page can therefore earn impressions for different reasons and at different stages of awareness.
Keep separate baselines, cohorts, and narratives:
- Search generative AI: evaluate topic demand, conventional query performance, page quality, intent satisfaction, and changes around search-facing content.
- Discover generative AI: evaluate freshness, visual presentation, audience affinity, editorial timing, and the page’s wider Discover behaviour.
Combining the two into one “AI visibility” number hides how users encountered the content. DMT’s detailed Search Console platform-property guide applies the same principle to social and video data: a useful report preserves the source surface instead of collapsing unlike journeys.
A practical measurement architecture
No single tool answers every AI-search question. Use four evidence layers.
Layer 1: Search Console generative AI exposure
Record which pages receive impressions, their trend, country/device distribution, and whether the activity appears in Search or Discover. This is first-party exposure evidence.
Layer 2: conventional Search Console demand and clicks
For the same page cohort and date range, analyse normal Search Console queries, impressions, clicks, CTR, and average position. This does not magically identify the hidden AI queries, but it reveals the page’s wider search-demand context and helps locate query families worth investigating.
Layer 3: GA4 landing-page behaviour
Use GA4 to examine sessions, engaged sessions, key events, and conversions for the same landing pages and dates. Compare against a prior period and a suitable control cohort. Be precise: a correlation between rising AI impressions and improving landing-page outcomes is a signal to investigate, not proof that AI visibility caused the change.
Layer 4: controlled answer observation
Maintain a small manual prompt log for commercially or reputationally important questions. Record prompt, surface, country, language, device, date/time, account or personalisation state, answer summary, cited domains, your URL if present, and a screenshot. Manual checks are samples, not population data, but they show answer quality that the report does not expose.
This combined method is more credible than treating any third-party “AI visibility score” as ground truth. It also fits the evidence-first practices in DMT’s generative engine optimisation guide.
Weekly Search Console generative AI workflow
Step 1: confirm scope and availability
- Record the Search Console property and whether it is a domain or URL-prefix property.
- Confirm the user has access to the dedicated Search and/or Discover report.
- Record the export date and the timezone used by your broader reporting system.
- Do not interpret an absent report as zero market visibility; rollout is staged.
Step 2: export the report before interpreting it
Export page, country, device, and date data for a consistent weekly or 28-day window. Retain the raw export. Search Console interfaces and retention windows can change; the raw snapshot is your evidence receipt.
Step 3: create page cohorts
Group URLs by a business-relevant dimension instead of reviewing an unstructured list:
- topic cluster or entity;
- content type such as guide, comparison, category, product, or tool;
- funnel role;
- publish or refresh month;
- author or subject-matter owner;
- template;
- market and language.
Cohorts make the report actionable. If comparison pages gain exposure while shallow announcements do not, that is a stronger editorial clue than “AI impressions increased 12%.”
Step 4: calculate useful derived metrics
The report does not provide clicks or CTR, but you can still calculate descriptive measures without pretending they are conversion metrics:
- AI impression growth: current period versus previous comparable period;
- page concentration: share of impressions held by the top 10 pages;
- newly visible pages: URLs with first observed impressions in the period;
- lost-visibility pages: previously visible URLs that fall to zero or materially decline;
- market concentration: country mix and changes;
- device mix: mobile, desktop, and other device changes;
- cohort lift: change for refreshed pages versus a stable comparison cohort.
Use median page change as well as totals. A site-wide total can rise because one URL spiked while the rest of the content weakened.
Step 5: reconcile with conventional search and GA4
Join data on canonical landing-page URL and consistent dates. Normalise protocol, hostname, trailing slash, parameters, and redirects before joining. Then ask:
- Did generative AI exposure rise while overall search impressions also rose?
- Did conventional clicks change for the same pages?
- Did organic landing sessions or engaged sessions change?
- Did the page cohort change, or only one outlier?
- Was there a content refresh, technical release, seasonality event, or algorithm update at the same time?
A page can gain AI exposure without gaining clicks, especially when the generated result satisfies the question. That is not automatically a failure. The business question is whether the visibility improves qualified discovery, brand demand, assisted journeys, or downstream outcomes. The report cannot answer that alone.
Step 6: inspect a small priority sample
Select high-impression pages, sudden winners, sudden losers, and decision-stage pages. Review their conventional queries, manual prompt sample, cited competitors, content freshness, source quality, and technical accessibility. Avoid checking hundreds of prompts without a sampling plan; personalisation and result volatility make casual spot checks noisy.
Step 7: choose one action and one validation
Every recommendation should include the next piece of evidence that would confirm or reject it. Examples:
- Observation: an updated comparison cluster gained AI impressions. Action: improve missing decision criteria on two related pages. Validation: compare the cohort with unedited pages for four weeks.
- Observation: a guide has high impressions but poor manual representation. Action: strengthen definitions, source citations, and explicit factual statements. Validation: recheck a fixed prompt sample and report trend.
- Observation: exposure is concentrated in one country. Action: do not immediately localise. Validation: confirm conventional demand, audience value, and content gaps in the proposed market first.
Dashboard schema you can implement
| Field | Source | Decision use |
|---|---|---|
| Date, surface, page, country, device, AI impressions | GSC generative AI report | Exposure trend and distribution |
| Query, page, clicks, impressions, CTR, position | Conventional GSC performance | Demand and click context |
| Sessions, engaged sessions, key events, conversions | GA4 landing-page report | Post-click outcome context |
| Cluster, content type, publish date, last refresh | CMS/content inventory | Cohort comparison and recency |
| Prompt, answer, citations, context, screenshot | Controlled manual log | Representation and citation quality |
| Change, hypothesis, owner, validation date | Experiment ledger | Accountability and learning |
Keep raw data and interpretation separate. A dashboard should let another analyst trace a conclusion back to an export, page, date, and calculation.
The reporting layer should also feed editorial priorities rather than sit in an analytics silo. Use the cohort evidence with DMT’s SEO content strategy framework to decide whether a page needs a factual refresh, supporting article, internal-link improvement, consolidation, or no change at all.
What not to conclude from the report
- “AI impressions increased, so AI traffic increased.” The dedicated report does not show clicks.
- “This page generated revenue through AI Mode.” It does not provide conversion attribution.
- “This keyword triggered our impression.” Queries are not shown in the dedicated report.
- “Google quoted this exact passage.” The report does not expose the generated answer.
- “The report is missing, so our brand has no AI visibility.” Rollout is staged and Search Labs is excluded.
- “We need special AI schema.” Google says existing SEO fundamentals apply; there is no special AI file or markup required to appear in these features.
- “More impressions always mean better performance.” Exposure can be irrelevant, inaccurate, or disconnected from business value.
How to improve visibility without chasing an AI-only trick
Google’s AI features optimisation guidance emphasises the same foundations used in Search: create helpful, reliable, people-first content; ensure Google can crawl and index it; make important content available in text; support the page with a good experience; and keep structured data consistent with visible content. There is no magic generative-AI schema.
For publishers, the durable priorities are:
- Answer the real decision: include definitions, comparisons, constraints, exceptions, examples, and the next action.
- Show evidence: cite primary sources, name the fact-check date, and distinguish fact from interpretation.
- Close information debt: update stale prices, product states, regulations, screenshots, and recommendations.
- Build entity and topic coherence: connect supporting pages with descriptive internal links and avoid thin duplicates.
- Make pages technically accessible: preserve indexability, canonical consistency, useful status codes, and renderable main content.
- Improve the reader experience: use concise direct answers, scannable structure, original analysis, and clear provenance.
DMT’s AI Overviews SEO guide covers these content and technical controls in more depth. The broader AI and SEO guide explains why citation eligibility, brand authority, and conventional rankings should be analysed together.
Opting out and controlling previews
If a publisher does not want content used in particular Search result treatments, use Google’s documented search preview controls and validate the result carefully. Controls such as snippet restrictions can also reduce the visibility and usefulness of ordinary search results. This is a product, legal, traffic, and brand decision – not a setting an SEO team should change casually.
Before changing controls:
- identify the exact content and surface in scope;
- read the current Google documentation rather than relying on an old screenshot;
- model the effect on normal Search as well as generative features;
- obtain the appropriate legal and business approval;
- test on a limited page cohort where possible;
- monitor crawl, indexing, snippets, impressions, and clicks after release.
Why a report may show no data
Check these possibilities before declaring a technical problem:
- the report has not rolled out to the property;
- the selected dates contain no reportable impressions;
- filters exclude the available data;
- the user lacks sufficient property access;
- the site or page is not eligible, indexed, or visible for the relevant experiences;
- the observed activity occurred in Search Labs, which Google says is excluded;
- the report is being compared with a third-party metric that uses a different definition.
Keep a screenshot and timestamp when reporting availability. “I cannot see the report” is not the same fact as “the property has zero generative AI impressions.”
A 30-minute first analysis
- Export the last available 28 days from the Search report and Discover report separately.
- List the top 20 pages by AI impressions and the top 10 gainers and decliners.
- Add cluster, content type, publish date, and last-refresh date.
- Pull conventional GSC and GA4 context for those pages over the same dates.
- Select five priority pages for a controlled manual prompt check.
- Write observations, plausible explanations, missing evidence, and one reversible test.
- Schedule the same export next week; do not redesign the content programme from one snapshot.
The purpose is not to manufacture a new vanity score. It is to create a repeatable evidence trail from exposure to investigation to action.
Frequently asked questions
Does Search Console show clicks from Google’s generative AI results?
The dedicated generative AI performance reports do not currently expose clicks or CTR. Their impressions are included in overall performance data, but the dedicated view should not be treated as a click-attribution report.
Can I see the prompts or queries that triggered an AI impression?
Not in the dedicated report. Use conventional Search Console queries to understand the page’s broader search-demand context and a controlled manual prompt log to observe a small, clearly labelled sample.
Does the report cover both AI Mode and AI Overviews?
Google describes reporting for generative AI experiences within Search and a separate report for Discover. Check the current official help page for the exact included experiences because product naming and coverage can change.
Why do I not see the generative AI report?
Google is rolling it out in stages to a subset of sites. Access, data volume, dates, filters, and property eligibility can also affect what appears. Search Labs data is excluded.
Do I need special schema to rank in AI Overviews or AI Mode?
No special AI schema is required. Google recommends established Search fundamentals and says structured data should accurately match visible content. Useful schema can help Google understand a page, but it is not a generative-AI access switch.
Can I connect AI impressions to GA4 conversions?
You can compare page cohorts and dates, but the dedicated report does not expose the user-level click path needed for direct conversion attribution. Treat correlations as investigation signals and use controlled tests, other first-party data, and honest uncertainty.
Editorial note: Google is actively evolving generative AI experiences and reporting. The official sources linked above were checked on 17 July 2026. Reconfirm definitions, availability, and controls before making a material decision.
