Can Google Search Console Measure Your AI Visibility?

Google will now tell you how often your site appears in its AI answers. The generative AI performance report, announced on 3 June 2026 and rolled out to properties worldwide on 31 August, gives Search Console a dedicated view of impressions in AI Overviews and AI Mode. It is the first first-party number most teams have ever had for AI visibility on Google — and it answers about a third of the question teams are actually asking.

Key takeaways

Google's help documentation is short, and worth taking literally. The report shows one metric: "Impressions are how many times links to your site were shown to a user in a generative AI feature on Google Search." It covers two features under one number — "AI Overviews, AI Mode" — and offers four dimensions.

A second report covers generative AI features in Discover, and the same June announcement shipped the opt-out toggle — the Search generative AI control — which we covered separately in should you opt out of AI Overviews. This post is about the measurement, because that's the half teams will use weekly.

One more line from the documentation deserves attention: the report "includes data from the Web search type in the Performance report." Your AI impressions are not new impressions. They were being counted inside your ordinary Web totals the whole time; the new report labels a slice you previously couldn't see. Two things follow. First, nothing about your historical Performance report changed on 31 August — if impressions rose while clicks stayed flat over the past year, AI features were already part of that story. Second, the tempting arithmetic — subtract the AI report from the Web report to get "classic-only" performance — is only approximate: the AI view is impressions-only, its newest data is documented as preliminary, and the two reports don't share every dimension.

No queries. The report tells you page 12 appeared in AI answers 400 times, not what anyone asked to make that happen. Content work starts from the question — which prompts you're surfacing for, which phrasings retrieve a competitor instead — and none of that is here. Microsoft's equivalent shows its retrieval phrasings (grounding queries in Clarity); Google shows URLs. The only way to attach questions to your Google AI presence today is to ask the questions yourself and record the answers, which is prompt tracking.

No clicks. An impression tells you exposure, not value. The best public estimate of the gap is still the Pew Research Center's click study: in roughly 69,000 observed searches, users clicked a link inside an AI summary in about 1% of visits where one appeared. So read the impression count the way you'd read a billboard audit — how often you were shown, to whom, where — rather than as a traffic forecast. If being named in the answer is the outcome your business needs (software, B2B, services), that reading is fine; the number is genuinely useful. If you need the visit, this report can't tell you whether you're getting it.

No split between AI Overviews and AI Mode. A link under a summary at the top of an otherwise classic results page and a link inside a fully conversational AI Mode session are different events with different user intents, and the report adds them together. As AI Mode grows, the blended number gets harder to interpret: a rising trend could be more summaries above classic results, or more of your buyers moving their entire session into a chat surface. The report won't tell you which.

There's a subtler absence behind all three: the report counts links shown, not brands discussed. An AI answer can describe your product — accurately or not — without linking to you at all, a pattern we've written about as ghost citations. A brand with weak link presence and strong model-knowledge presence will look invisible in this report while appearing, unlinked, in answer after answer. Impressions of links are the measurable shadow of your AI presence, not the presence itself.

Six months ago the honest answer to "can we measure AI visibility with first-party data?" was no. Now there are two platform-native sources plus sampling, and they answer different questions. The mistake to avoid is treating any one of them as the AI visibility number.

The two first-party reports are precise and narrow: observed data, no sampling noise, one ecosystem each, silent on wording. Sampling is the inverse: engine-agnostic and answer-aware, but an estimate — and a noisy one, which is why we keep pointing at our citation volatility data showing 60–70% day-to-day turnover in cited domains across tracked prompts. A single manual check of ChatGPT was never a measurement; a Search Console impression count is a real one, for the slice it sees.

Can Search Console measure your AI visibility? It can now measure one honest, narrow thing: how often Google's own AI features showed links to your site. That's worth having, worth baselining, and worth reading weekly. What it structurally cannot do — name the questions, value the exposure, see the wording, or see any engine but Google — defines the rest of the measurement stack, not a reason to dismiss the report.

Elmo is an open-source, self-hosted AI visibility platform that covers the other column of the table: it runs your prompt sets across ChatGPT, Claude, Gemini, Perplexity, and Google's AI surfaces and records every answer, mention, citation, and competitor named beside you — the wording and the who-else that no platform's first-party report will ever show you.

For the fundamentals, start with what AI citations are and how to track your brand in AI search. For the adjacent first-party source, see Microsoft Clarity's AI citations; for the toggle that shipped alongside this report, should you opt out of AI Overviews. For the vocabulary, the AI search glossary.