Turn AI Search Impressions Into a Useful Publisher Report

Google Search Console’s generative AI performance report shows where and when a site appears in supported AI search features. Publishers can use it to measure exposure, but not visits or business value: the report does not currently provide clicks, click-through rate, position or queries.

Google launched separate Search and Discover reports on June 3, 2026. The Search report covers impressions in features such as AI Overviews and AI Mode; the Discover report covers generative AI features in Discover (Google Search Central). Google says the reports completed their worldwide rollout on August 31, 2026 (Search Console Help).

Enter two comparable impression totals; add top-page impressions to measure concentration.

The Search Report Measures Exposure

The report’s current metric is impressions: the number of times links to the site appeared to a user in a supported generative AI feature. Operators can break those impressions down by:

• page;
• country;
• date, grouped by day, week or month; and
• device: desktop, tablet or mobile.

The Search report also provides filters for text-based and multimodal web search. Its page dimension usually assigns data to the canonical URL after redirects rather than to a duplicate URL. The same 1,000-row limit and other general limitations as the standard Search performance report apply (Search Console Help).

If unexpected URLs dominate the page breakdown, review consolidation before treating every row as a distinct content result. Our guide to canonical tags and noindex explains the relevant controls.

As of September 27, 2026, Google documents no click, click-through-rate, position or query dimension in this report. An impression increase means the site appeared more often. It does not establish that more people visited, subscribed or bought anything.

Preserve Comparable Raw Exports

Export the report on a fixed weekly or monthly schedule. Keep each raw export with the following context:

• Search Console property;
• extraction date;
• selected reporting period;
• Search or Discover report;
• search type; and
• active filters.

The newest data points may be preliminary and can change as Google finishes collecting data. Search Console also warns that chart totals can differ from table totals because the chart may aggregate by property while page rows aggregate by URL (Search Console Help). Compare like with like rather than forcing every displayed total to reconcile.

A period-over-period calculation is only meaningful when both exports cover comparable durations, properties and filters. A seven-day mobile-only total should not be compared with a full month across all devices without clearly separating the change in scope from the change in visibility.

Weekly groupings are generally easier to interpret than daily movement when impression volume is low. Monthly exports may be more suitable for properties with sparse data, provided the longer interval does not obscure a publishing or technical change.

Report Concentration Alongside Growth

A property-level increase can conceal a declining priority section while a large archive grows. Start with three breakdowns:

1. Pages: Identify which canonical URLs account for most AI impressions.
2. Countries: Check whether visibility is concentrated in markets the publication serves.
3. Devices: Determine whether a change is broad or isolated to desktop, tablet or mobile.

Track a defined cohort of commercially or editorially important pages separately from the whole property. For example, the sitewide total and a buying-guide cohort answer different questions even when both use the same impression metric.

A compact operating table is usually sufficient:

Field	Purpose
Weekly AI impressions	Directional visibility trend
Top pages and share	Concentration and dependency
Priority-page cohort	Visibility for important URLs
Country and device split	Audience relevance and distribution shifts
Release notes	Publishing, template and technical changes

The share attributable to top pages is particularly useful. Rising impressions concentrated in one URL represent a different operating condition from the same growth spread across a large cohort. Concentration is not inherently good or bad, but it exposes dependency that a sitewide total hides.

Keep AI Impressions Separate From Analytics Outcomes

Use site analytics to inspect organic entrances, engagement and conversions for the same page cohort and period, but keep the datasets separate. Without click-level attribution in the generative AI report, concurrent movement is not proof that AI impressions caused visits.

If a group of buying guides gains AI impressions and organic entrances in the same week, record the result as a correlation worth investigating. Do not label the entrance increase “AI traffic” unless another reliable source identifies it as such.

The same restriction applies to engagement and revenue. A conversion recorded on a page that also gained AI impressions does not connect the conversion to an AI Overview or AI Mode appearance. Search Console supplies evidence of exposure; analytics supplies evidence of activity on the site. The available fields do not provide the missing attribution link.

This distinction should survive into dashboards and executive summaries. Label the Search Console series “generative AI impressions,” not “AI visits,” “AI referrals” or “AI conversions.” If analytics is displayed beside it, describe any shared movement as correlated rather than attributed.

Annotate Changes Before Evaluating Them

Record publication dates, substantial revisions, canonical changes, migrations and outages. Define the expected mechanism before reviewing the result. A proposed hypothesis might be: “Improved source documentation will increase impressions for this 20-page cohort over eight weeks.”

Use a comparison cohort where practical and choose the evaluation window in advance. The content experiment design framework provides a reusable structure for hypotheses, metrics, guardrails and stopping rules.

An intervention log helps separate plausible effects from unrelated changes. If impressions move after a template release, record whether the release altered canonicalization, crawlable content, structured data or page availability. The report cannot identify the cause on its own.

Do not respond to every fluctuation by producing pages for imagined query variants. Google says its generative AI features use core Search ranking and quality systems, and advises publishers to create useful, non-commodity content rather than scaled pages aimed at fan-out queries. It also says no special AI markup, llms.txt file or AI-specific writing style is required (Google’s generative AI search guidance).

A Missing Report Usually Means Insufficient Data

Worldwide availability does not mean every property will display data. Google says a property may not show the report when it has not accumulated enough generative AI impressions. Verify that the relevant Search Console property is selected and that the site remains included in Search generative AI features (Search Console Help).

The inclusion setting is under Settings → Search generative AI. Excluding a property prevents its links and content from appearing in AI Overviews, AI Mode and generative AI features in Discover, so it will receive no impressions or traffic from those surfaces.

Google says the setting does not affect ranking or inclusion elsewhere in Search and does not control AI training. Google-Extended is the separate training control (Search Console Help).

Treat the inclusion switch as a distribution policy decision rather than a reporting optimization. First establish what visibility the property receives, then weigh that exposure against the publication’s content-use policy. The performance report can inform that decision, but without click and outcome metrics it cannot calculate the full value of participation.