Google Search Console’s AI report: what the numbers mean.

01 / WHAT CHANGED

What does Google’s AI report measure?

Google announced dedicated generative AI performance reports on 3 June 2026. The Search version covers AI Overviews and AI Mode; Discover has its own report. Google’s announcement lists impressions, pages, countries, devices for Search, and dates. The underlying data remains included in the overall performance report. [2]

The current Search documentation does not list dedicated clicks, CTR, position or query data. Treat those as reporting limits, not blanks to fill with estimates. It also describes a phased rollout and says insufficient impressions can explain a missing report. An unavailable view is not a measured zero. [1]

GOOGLE’S OFFICIAL SCREEN EXAMPLE

Where the numbers appear

Open the full-resolution image ↗

Start with the scope. The sidebar separates the Search and Discover AI reports.

Then check the period. The date filter sits above the impressions card and chart.

Finally, check the grouping. Pages, Countries, Devices and Dates appear below the chart. Selecting a different grouping changes what the table describes.

MEW’s recommendation: decide what question each metric answers before putting it on a dashboard. An impression can help you identify a page to investigate. Whether people reached that page, understood the offer or contacted the business requires other evidence.

Do not add AI impressions to an overall Search total that already contains them. Nor should you call this a measure of visibility across every AI service: the scope here is Google’s supported features, not ChatGPT, Perplexity or the whole answer-engine market. [2]

02 / THE COUNTING RULE

Why can the page table exceed the chart?

They can count different units. Google distinguishes aggregation by property from aggregation by page. Multiple results from one property count once at property level; distinct URLs are counted separately at page level. [3]

In the AI report, the unfiltered chart is aggregated by property. Adding a URL filter changes its aggregation to URL level. The table’s aggregation depends on the selected dimension. [1]

Before explaining a discrepancy, record the filters and grouping. A page-total increase might reflect more pages appearing together, rather than more separate appearances for the property. That is an interpretation to investigate, not something the example proves about your site.

Before comparing totals, ask “What did each total count?” Do not force unlike totals to reconcile by deleting rows or changing the period.

03 / JAPAN AND ENGLISH PAGES

Does country data tell you which language works?

No. Google’s country dimension identifies where a search originated. Page reporting generally assigns data to the canonical URL. Neither tells you the language of the page without a separate classification. [4]

MEW’s approach: keep a small URL register with the page’s actual language, intended market and purpose. Match the reported canonical URL to that register. An English page viewed from Japan remains an English page; a Japanese page viewed abroad remains Japanese. Leave ambiguous or unmatched URLs unclassified until checked.

Start with two separate views: which pages appear, and which countries account for visibility. Do not invent a page-by-country breakdown by joining two aggregate exports. Use that combination only when a supported report or filter actually supplies it, and preserve the filter conditions.

Compare similar page purposes, devices and complete periods before changing editorial resources. One newly published Japanese guide and an established English tool page are not a controlled language test. Even a sustained difference would be an observation under those conditions, not proof that translation or a familiar visual style caused it.

04 / A REPORT SOMEONE CAN CHECK

What should an AI visibility report include?

A useful report makes its numbers traceable. Keep the measurement conditions beside the finding rather than burying them in a separate file. The checklist below is MEW’s proposed reporting practice, not a Google-required template.

There is a practical export trap: Google says values shown as ~ or - can become zeros in downloaded data. Its data guide also describes preliminary values that may change. Preserve the original state before building comparisons or charts. [3]

Use the report to choose what to inspect, then test a specific improvement. If a page appears but buyers cannot verify pricing, support or product fit, more visibility alone does not repair that gap. Our guide to AI discovery and Japanese buyer evidence examines what the destination page needs to establish.

Keep visibility, visits and qualified enquiries separate. A useful next action can connect them; a single score cannot explain them.