Google's Generative AI Report: A Field Guide

For two years the standard complaint about AI search was that you could not measure it. You could see traffic falling and guess why. Now there is a first-party number, free, in a tool you already have. It is limited — impressions only, no clicks or queries — but it is the first official window into whether AI surfaces are using your content.

This is a walkthrough of what the report shows, what it does not, and how to read it. We have used this site's own data throughout, because a worked example beats a description.

Finding it

In Search Console, open Performance, then look for the Generative AI report link beneath the main chart. It is marked Beta. The default view is a three-month window.

You get impressions — how often your pages appeared in AI Overviews or AI Mode — broken down by page, country and device. You do not get clicks, you do not get queries, and the data is not backfilled far. Treat it as directional.

Reading it: a worked example

Over a recent three-month window this site recorded 411 impressions across AI surfaces. That is a small number, and the absolute figure is not the interesting part. The distribution is.

The top pages were, in order: the site's definitional page for its core term (139 impressions), the hub page for its second discipline (113), the homepage and its main section landing page (105 between them), and then one blog post (31) and a long tail of reference-style pages — glossary, playbook, audit checklist.

Two things fall straight out of that shape.

AI surfaces cite definitions. The pages doing the work are the ones that answer "what is X?" for a specific term, in a self-contained paragraph. Nothing promotional appears anywhere in the list. If you want a testable prediction from your own data, this is it: look at which of your pages appear, and you will probably find they are the ones a model could quote a sentence from without needing the rest of the page.

One blog post appeared and four did not. The one that surfaced was the post built on a named, dated study with real numbers. The posts built on general argument did not register at all. That is a single site's data and not proof of anything on its own — but it points the same direction as the published research on citation behavior, and it is cheap to check against your own content.

The trend line matters more than the total

The most useful thing in the report is not the number, it is the shape over time.

On this site, daily AI impressions ran between four and ten through May, June and early July, spiked to eighteen in mid-July, then decayed toward zero across August. The cause is not mysterious: the site had not published or meaningfully updated anything since late 2025, and every freshness signal it emitted — sitemap timestamps, dateline text, machine-readable policy files — was frozen at the same old date.

That decay is consistent with what the broader research shows. An analysis of millions of AI citations in 2026 found cited URLs were meaningfully fresher than the web at large. Freshness appears to matter more to AI surfaces than it does to classic search — which stands to reason, since an answer engine staking its credibility on a claim has more reason to prefer recent sources than a ranking system does.

If your Generative AI line is drifting downward while your ordinary Search line holds steady, stale content is the first thing to check.

What the report cannot tell you

Be honest about the limits before you build reporting on it.

• No queries. You cannot see what prompt surfaced you, so you cannot do keyword work against it.
• No clicks. Impressions are not visits. AI referral traffic remains a small share of most sites' sessions even where impressions look healthy.
• Google only. ChatGPT, Perplexity, Claude and Copilot are all invisible here, and they diverge sharply — one 2026 analysis found only about 11% of domains cited by both ChatGPT and Perplexity overlapped. A Google-only measure is not an AI-visibility measure.
• Attribution is leaky elsewhere. A large share of AI-referred sessions arrive without a referrer and land in "Direct" in analytics, so GA4 undercounts AI traffic substantially.

The other half of the release: the opt-out control

Alongside the report, Google shipped a control to exclude your content from AI Overviews, AI Mode and Discover AI features without affecting normal Search rankings. Previously this was an all-or-nothing choice via Google-Extended, which is why so few publishers used it.

Whether to use it is a real strategic question, not a technical one. Blocking protects against substitution — the case where an AI answer satisfies the user completely and no visit follows. Allowing keeps you eligible for citation and the brand presence that comes with it. Roughly 68% of US Google searches already end without a click, so the substitution is happening whether or not you are cited; the question is whether you would rather be quoted or absent.

Most publishers should leave it open and compete on being the best source. Sites whose entire product is the information itself — where the answer is the business — have a genuine case for blocking.

What to do with this

The honest summary: this report is the best free signal available, and it is still a narrow one. Use it to confirm direction, not to prove ROI.

References

• Google Search Central, "Introducing generative AI performance reports", June 3, 2026
• Ahrefs, "AI search overlap", 2026
• SparkToro, "In 2026, less than one third of Google searches still send a click", 2026