Google added a dedicated generative AI performance report to Search Console in June 2026. It shows which of your pages appeared in AI Overviews and AI Mode, broken down by date, country and device. That is real, usable evidence—but it is easy to over-read. The most common mistake is treating an impression as proof that an article was quoted, trusted or responsible for a sale.

The report is still rolling out to a subset of properties, so you may not have it yet. If you do, use it to find patterns worth investigating without asking the chart to support a story it cannot prove.

First, understand what the report counts

An impression is recorded when a link to your site appears in a supported generative AI feature—currently AI Overviews and AI Mode. Search Labs experiments do not count because they are still in active development.

The chart totals and table totals will not always match, and that is expected rather than a bug. The chart aggregates by property, so two links from the same site in one generative response can collapse into a single impression. Filter to a specific URL and the aggregation changes to the page level.

Most data is assigned to the canonical URL after redirects. A duplicate URL can therefore fold into the canonical page’s row instead of appearing separately.

What it will not tell you

The current report does not provide a universal answer to these questions:

  • Which exact query triggered every appearance?
  • Which sentence or claim from the page influenced the response?
  • Was the link visually prominent?
  • Did the user notice or trust it?
  • Did the impression produce a visit or conversion?

Google’s description sticks to impressions, pages, countries, devices and dates. Read those as observation points worth investigating, not as proof of editorial authority. The report was not built to make that case.

A four-layer way to use it

1. Visibility: which pages appear?

Export the page table and group URLs by content type: guide, review, comparison, category page or commercial landing page. Calculate each group’s share of generative AI impressions.

If a product comparison appears more often than a broad category page, that is a reasonable hint that decision-focused content fits complex AI answers better than a general overview. It is not proof of causation. Record it as a hypothesis and see whether the pattern holds.

2. Stability: is the pattern repeatable?

Compare at least two equal periods and exclude the newest preliminary data. A single spike can reflect a temporary topic, a small denominator or a reporting change.

Keep a short annotation log alongside the export:

Date Page or group Change Why it may matter
8 Aug Comparison pages Added decision tables Easier extraction of explicit trade-offs
15 Aug Three guides Updated primary sources Time-sensitive claims became current

Do not rewrite an article the moment a number moves. Wait for a pattern that appears across several weeks and more than one page.

3. Engagement: what happens after visibility?

The generative AI report centres on impressions. Combine it with the normal Search performance report and analytics data. For the same landing-page group, watch:

  • organic clicks and click-through rate;
  • engaged reading or scroll depth;
  • newsletter sign-ups;
  • outbound affiliate clicks;
  • assisted conversions, where your analytics setup supports them.

A page can gain generative visibility without gaining clicks. Brand exposure may still have value, but it is not qualified traffic until a visitor arrives and completes a useful action.

4. Editorial diagnosis: why might these pages be selected?

Inspect the strongest and weakest pages using the same checklist. Look for a direct answer near the top, primary sources, explicit comparison criteria, current dates, useful tables and a clear relationship to the rest of the topic cluster.

The goal is not to copy the surface layout of the winning page. Identify the editorial work that makes its claims easier to verify and its purpose easier to grasp. That is the part worth repeating elsewhere.

A monthly scorecard worth keeping

Use one row per content group rather than pretending every URL has enough data for a reliable conclusion.

Content group AI impressions Change vs prior period Organic clicks Useful action rate Decision
Comparisons Maintain / investigate / update
Reviews Maintain / investigate / update
Practical guides Maintain / investigate / update

Add a note for data limitations. Search Console exports may contain zeros where the interface displays unavailable values, and the usual row limits and aggregation differences still apply.

If the report is missing

Access is still limited during rollout, so that is the first explanation to rule out. A property may also have too few impressions or may have excluded itself from Search generative AI features. Confirm that important pages are indexable and eligible for snippets, but do not make technical changes merely to force the report to appear.

Continue measuring normal Search performance and business outcomes. Google states that established SEO requirements remain the foundation for AI features; there is no separate schema or machine-readable AI file required.

What this report can—and cannot—decide

Use the report to decide where deeper analysis is justified. It can show that a group of pages is gaining or losing visibility in supported Google AI features. It cannot tell you, on its own, what to write next or whether a page created revenue.

The workflow that holds up under scrutiny is modest: export, group, compare, annotate and connect visibility with reader behaviour. It will not produce a dramatic headline for a team update, but it produces decisions you can actually defend.

Sources