Google's Generative AI performance report is now live for every Search Console property, and for the first time it isolates the impressions your pages earn inside AI Overviews and AI Mode from the classic blue-link data. That closes a two-year measurement gap. It also stops well short of answering the question most teams actually have. The report shows impressions only: no clicks, no click-through rate, no queries, no positions, no split between AI Overviews and AI Mode, and no API access. Read correctly, it is a useful presence signal. Read loosely, it will be turned into "AI Overview rankings" that do not exist.
Here is what the report contains, what it leaves out, and a routine for using it without fooling yourself.
What the report shows
Google announced the report on June 3, 2026, rolled it to a subset of sites for testing, and completed the rollout to all properties around August 11. It sits under Performance and Generative AI, still carries a Beta label, and counts an impression whenever one of your URLs appears inside an AI Overview or an AI Mode response. You can break impressions down by page, country, device, and date, and a separate view covers generative features in Discover.
One detail is worth internalizing before you report anything to anyone: these impressions are not new traffic on top of your existing numbers. Google's documentation is explicit that the Generative AI report draws from the Web search type of the existing Performance report. The impressions were always being counted; they have been separated into their own view. Your totals did not change.
What the report hides
The gaps matter more than the contents.
No clicks or CTR. You can see that a page appeared inside an AI answer. You cannot see whether anyone clicked it from there. Google has said more metrics will come over time, without a timeline.
No queries. The report withholds the prompts that triggered the impression, so you cannot tell which questions your pages are being pulled into.
No positions, and no split between surfaces. An impression from an AI Overview and an impression from AI Mode are counted together, even though the two are different products with different click behavior. If you have not yet separated them in your own thinking, our AI Mode versus AI Overviews explainer covers why they diverge.
No API. As of mid August, the Search Console API rejects generative AI values for the search type parameter; only web, image, video, news, Discover, and Google News are accepted. You are reading this in the interface or not at all.
And a reliability caveat for anyone who pulled the report in its first fortnight: from around August 12 to August 20, a confirmed bug showed widespread drops in impressions across both the standard and Generative AI reports before recovering. If your August trend line has a cliff in it, that is the bug, not your visibility.
How to read it without fooling yourself
Three rules cover most of the mistakes.
An impression is a citation, not a mention. The report counts your URL being used as a source. It says nothing about whether the answer named your brand, recommended it, described it correctly, or handed the recommendation to a competitor while citing your page for a definition. Those are the outcomes that move buyers, and the report is silent on all of them.
Treat AI impressions as presence, not traffic. Early independent data suggests how weak the impression-to-click link is on these surfaces. One site owner who pulled 90 days of Search Console data on August 12 and classified prompt-shaped queries (eight or more words, or ending in a question mark) found that 524 of 527 such query-page pairs returned zero clicks across 6,139 impressions, and that a top-10 position for a prompt-shaped query was worth roughly one fourteenth of a top-10 position for a keyword. It is one site, but it matches everything else we know about click behavior inside AI answers. Do not build a traffic forecast on this report.
Use it for page-level inclusion and trend. What the report is good at is showing which of your pages Google's AI features actually pull from, and whether that set is growing or shrinking. That is useful input for deciding which pages to refresh, which to structure better, and which are doing quiet work you never credited.
What it cannot tell you that decides outcomes
The report is, by design, a Google instrument for Google surfaces. It cannot tell you what the answer said, whether you were recommended, how the description compared with reality, which competitors were named, or how your share of those answers is trending. And it covers two of the AI surfaces your buyers use. ChatGPT, Perplexity, Claude, Copilot, Gemini's app, Grok, and the open-weight models have no Search Console equivalent and never will.
That is the layer answer-level tracking exists for. Diploria polls the questions your buyers ask across 11 AI surfaces, including AI Overviews and AI Mode tracked separately, stores every full answer and citation, and reports mention rate, position, sentiment, and share of voice against named competitors. The Search Console report tells you your page appeared. Answer-level tracking tells you what happened next inside the answer.
A simple monthly routine
Pull the Generative AI report by page and note the top pages and the month-on-month direction. Compare against the same pages' organic clicks to spot pages that appear in AI features but are losing clicks; those are candidates for the decision covered in our piece on Google's new AI opt-out, and more often candidates for a refresh. Segment AI referral traffic in GA4 so you know what the surfaces send back, using the context in our AI referral traffic analysis to keep the numbers in proportion. Then check answer-level tracking for the same period: were you named, was it accurate, and who took the recommendation when you were not? The free AI visibility check is a fine starting point for that last step.
Four data points, one hour a month, and you will know more about your AI presence than most teams learn from a year of guessing.