What Google's AI performance report actually measures

The report documentation says it is being released to a subset of website owners for testing before a broader rollout. A missing report does not automatically mean a site has no AI visibility; access can depend on rollout status and whether the property has received enough impressions from the supported features.

That means teams should record when access becomes available and avoid comparing a property with the report to one without it as if the two dashboards were measuring the same thing.

The current Search report includes impressions from Google's AI Overviews and AI Mode. Google says those features can use query fan-out, where related searches are issued across subtopics and data sources before supporting links are selected.

The result is not a fixed list of websites that always wins. The links can vary by query, model, device, country, and the supporting pages Google identifies for that question.

The report can group data by pages, countries, dates, and devices. For pages, Google says the data is grouped by the final URL linked from a generative AI feature after redirects, and most performance data is assigned to the canonical URL rather than a duplicate URL.

This is another reason to keep article URLs stable. A content system that creates multiple versions of the same post makes reporting harder and can split attention between URLs that should represent one page.

An impression means a link to your site was shown to a user in a generative AI feature. It does not mean the user clicked, read the page, signed up, or remembered the brand. Use the Search Console performance documentation to understand the broader reporting model, then connect exposure to clicks and analytics conversions.

Google also notes that chart totals and table totals can differ because the chart is aggregated by property while some table views are aggregated by page. Small discrepancies are not automatically tracking failures.

Google marks the newest data as preliminary when it is still being collected. A dotted line in the chart indicates that the latest values may change over the next few hours.

Do not use an incomplete current day to declare that a topic has won or lost. Weekly comparisons and stable date ranges are less exciting, but they make a better operating view.

Start by exporting the pages and topics that receive AI impressions. Group them by search intent, then look for gaps between pages with exposure and pages that convert. A page with impressions but weak engagement may need a clearer promise, a better internal link, or a stronger next step.

Finally, compare changes in the report with changes in the underlying pages. This turns the report into a feedback loop for editorial decisions instead of a scoreboard that encourages random rewrites.

Use Google Trends before publishing to form a demand hypothesis: compare the broad topic with the specific question, inspect related searches, and check whether the interest is seasonal or sustained. Use Search Console after publishing to see how your own URLs actually perform. Those tools answer different questions and should not be merged into one invented score. Google's Trends FAQ explains the limits of the signal; its Search Console report explains the product-specific measurement.

Google Trends is a normalized sample of search interest, not a count of searches, and Google says it excludes searches made by Google products and services such as AI Mode and AI Overviews. Search Console's generative AI report is therefore the relevant product for measuring those AI-feature impressions when your property has access to it.