Google finally reports AI visibility. It sees about a third of it.

For two years the standard answer to "how do we measure AI search" was a shrug and a spreadsheet. On June 3, 2026 that changed: Google shipped Search Generative AI performance reports in Search Console — a dedicated view of how often your pages appear inside its generative AI features. It is the first time Google has conceded, in a product, that appearing in an AI answer is a thing worth counting.

Turn it on. It is free, it is first-party, and nothing else gives you Google's own numbers. Then read the rest of this page, because the gap between what the report shows and what "AI visibility" means is wide enough to build a strategy in.

What the report actually gives you

Two reports, one for Search and one for Discover. The Search one covers AI Overviews and AI Mode; the Discover one covers generative features there. You get impressions — the number of times a link to your site appeared inside one of those experiences — sliced by page, country, device and date, with history backfilled to roughly May 18, 2026. Rollout has been staged rather than instant, starting with a subset of properties, so an empty report may mean "not yet" rather than "not cited".

That is genuinely useful. It tells you which of your pages the AI layer is pulling in at all, and it is measured by the party doing the pulling. Every third-party estimate in this category is a sample; this is a census of one surface.

The four blind spots

• No clicks and no CTR. The report counts appearances, not visits. Google has said more metrics are coming over time, with no date attached. Until then you cannot connect an impression to a session.
• No queries. You learn that a page was pulled into an AI answer. You never learn which question pulled it — which is the one detail that would tell you what to write next.
• No placement and no passage. Being named as the recommendation and being the fourth link under a paragraph about someone else both count as one impression. The report cannot tell those apart, and they are worth wildly different amounts.
• No other engines. ChatGPT, Perplexity, Claude, Copilot — none of it. Google reports on Google. For a lot of B2B categories that is the minority of the assistant traffic that matters.

Why impressions-only is a specific kind of misleading

An impression logs a link to your site appearing. But the entire design goal of an AI answer is that the user does not need the link. The persuasion happens in the sentence that names you — "for teams your size, most people use X" — and that sentence can appear with no citation at all, or with the citation pointing at a review site that mentioned you. Neither shows up in your report.

The report counts the times you were linked. The value is in the times you were named. Those are different numbers, and only one of them is in the dashboard.

So read it as a floor, not a measure. It is the part of your AI presence that happens to be link-shaped, on one company's surfaces, without the question attached.

The stack that covers the rest

No single tool reports this end to end in 2026, and any vendor claiming otherwise is selling you one layer as if it were all four. A setup that actually holds up has four:

• Google Search Console — impressions on AI Overviews, AI Mode and Discover. Your first-party baseline for Google surfaces.
• Bing Webmaster Tools — the equivalent view where Copilot draws from. Small audience, near-zero effort, and it is the only free window onto a non-Google engine.
• Your analytics, with an AI referral channel — chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com grouped as one named channel. This is where the clicks Search Console will not show you actually land.
• Prompt-level checks across engines — run the questions your buyers ask, in each assistant, on a fixed schedule, and record whether you were mentioned, whether you were cited, and who was recommended in your place. This is the only layer that captures the named-but-not-linked case, and the only one that names the competitor who took the slot.

What to do this week

• Open Search Console and check whether the generative AI reports have reached your property. If they have, export the current window now — you want a baseline before you change anything.
• Compare the pages appearing in AI features against the pages you assumed were your strongest. The mismatch is usually the most interesting thing in the report — and it is rarely your best-ranked page.
• Create the AI referral channel in your analytics today, even if it is thirty sessions a month. You cannot backfill a segment you never defined.
• Write down ten prompts a real buyer would type. Run them in ChatGPT, Perplexity and Gemini. Note who gets recommended. That takes an hour and tells you more than the impressions chart will this quarter.
• Set a fixed monthly refresh. AI surfaces change presentation without warning; a number you only look at after a bad quarter is not a measurement.

The honest read

Search Console's AI reports are a real improvement over nothing, and "over nothing" is the correct comparison — that was the previous state of the art. But a report that shows impressions, on Google's surfaces only, without the query, is roughly a third of the picture. Treat it as one input in a four-part stack, and be suspicious of any plan that treats a single number from a single vendor's surface as your AI visibility.

Questions, answered

What is the Search Console generative AI performance report?

Announced by Google on June 3, 2026, it is a pair of Search Console reports showing how often links to your site appeared inside Google's generative AI features — AI Overviews and AI Mode in Search, and generative features in Discover. Data is broken out by page, country, device and date, backfilled to around May 18, 2026.

Does Search Console show ChatGPT or Perplexity visibility?

No. The reports cover Google's own surfaces only. Nothing in Search Console reflects ChatGPT, Perplexity, Claude or Copilot. For those you need referral segmentation in your analytics plus prompt-level checks run directly against each assistant.

Why does the AI report show impressions but no clicks?

Google launched it impressions-only and has said additional metrics will follow over time without committing to a date. There is also no query data, no average position and no indication of where in the answer your link appeared, so an impression cannot currently be tied to a session or to the question that triggered it.

How should I measure AI visibility properly in 2026?

Combine four sources on a fixed monthly cadence: Search Console for Google AI surface impressions, Bing Webmaster Tools for the Copilot side, an AI referral channel in your analytics for actual clicks, and scheduled prompt checks across ChatGPT, Perplexity and Gemini for mentions, citations and which competitor is recommended instead of you.