For two years, marketers measured Google's AI surfaces the way sailors measured icebergs: by what hit them. Traffic fell, impressions wobbled, and nothing in Search Console would say whether your pages were inside the answers doing the absorbing. That era formally ended this year: the generative AI performance report began rolling out in June as a limited test and reached essentially all properties by September, opt-out control included.
It's genuinely useful, materially incomplete, and easy to misread. This guide covers all three: what the report actually contains, the counting rules that shape its numbers, the five tracks worth a standing place in your reporting, and the instruments you still need beside it.
What The Report Is
You'll find it inside Search Console's performance section as a dedicated view of Search's generative AI features: AI Overviews and AI Mode, plus generative AI features in Discover. The data is impressions your URLs earn inside those surfaces, sliceable by page, country, device, and date.
Note the two absences before anything else, because they define how to use it. There is no click data for the AI surfaces, and there is no query dimension. You can see that your comparison page appeared in AI features 4,000 times last month, from which countries and devices, and how that trended. You cannot see which questions triggered it or how many of those appearances produced a visit. Exposure, not outcomes; pages, not questions.
The Counting Rules That Shape Every Trend
Google's documentation is specific about when an impression counts, and the specifics matter more than they look, as the help documentation spells out.
A link must actually be shown. Your content influencing an answer isn't an impression; a link to your site appearing is. You can be paraphrased into an answer invisibly and register nothing.
In AI Overviews, in-view means in view. A link counts only once it's scrolled or expanded into view. So AIO impressions measure reader engagement with the Overview's source panel, not mere inclusion, and a rise can mean readers are digging into sources rather than Google showing you more.
In AI Mode, follow-ups are new queries. Each turn of a conversation counts separately, and links shown in later responses earn fresh impressions. Long conversations can multiply impressions in ways that have nothing to do with broader reach.
Together these rules mean the report's numbers blend your presence with user behavior around it. That's fine, as long as nobody in your Monday meeting reads the line as "how often Google included us," because it isn't quite that.
15 hours a week manually. Or 15 minutes with RankControl.
Track citations, monitor competitors, and fix content gaps across every AI search engine. Automatically.
The Five Tracks Worth Watching
Track one: AI share of impressions. The ratio of AI-feature impressions to total impressions, trended monthly. This is your first-party measure of how much of your visibility has migrated into answer surfaces, and its slope is the single most strategic number in the report.
Track two: page winners and losers. Which pages earn AI exposure, and which pages that dominate classic results never appear in answers. The second list is the actionable one; pages that rank but don't surface in AI features usually have an extraction problem, not a ranking problem.
Track three: your own before-and-afters. Date every meaningful content change, then read AI impressions around the dates. Because the report is page-level, it's actually well-suited to this: restructure a page's opening this week, watch its AI-feature exposure over the following weeks, and you have a feedback loop that didn't exist a year ago.
Track four: device and country splits. Mostly boring, occasionally a finding: markets where AI features under- or over-serve your pages, or a mobile/desktop gap that flags rendering differences worth a technical look.
Track four and a half, for Discover users: if Discover matters to your traffic mix, its generative AI impressions sit in the same reporting family and deserve the same share-of-total treatment, particularly for consumer-leaning content where Discover's AI surfaces have been expanding.
Track five: the absorption gap. AI impressions rising while total clicks fall, on stable rankings, is the cleanest first-party signature of answers absorbing your clicks. Chart the two lines together and you've replaced a quarter of industry-study arguing with your own data.
Setting Up Your First Month With It
The report rewards a small amount of ceremony in week one, because trends only mean something against a baseline you actually recorded. New pages only appear once they're indexed, and our guide to getting new articles indexed through Search Console covers that step.
Start by exporting the current state: total impressions, AI-feature impressions, and the top 25 pages by AI exposure, into a dated sheet. Compute the AI share number once by hand so everyone in the room agrees on the definition, AI-feature impressions divided by total impressions, and write the formula into the sheet where the next person can see it. Then annotate your content calendar backwards: mark the last quarter's significant publishes and restructures on the same timeline, because the report's history lets you do retroactive before-and-afters on day one, and teams routinely find their first insight there, a restructured page whose AI exposure quietly tripled weeks before anyone had the report to notice.
Then set the cadence: a monthly read of the five tracks, fifteen minutes, appended to the same sheet. Weekly reads of this particular report mostly harvest noise, since the in-view counting rules make short windows jumpy; monthly is where its signals resolve. The exception is the fortnight after a major content push, where a weekly glance at the affected pages is fair diagnostics rather than anxiety.
Last, decide who owns the annotation habit, because the report's whole value compounds through dated context. An impressions line without your change log is trivia; the same line with publish dates, restructure dates, and campaign dates marked is the cheapest experimentation platform your content team has ever been handed.
AI search traffic grew 835% this year. Is your content ready?
RankControl generates 26 content formats optimized for ChatGPT, Claude, and Perplexity. Published on your domain, matched to your brand.
The Control Next Door
The rollout shipped with a companion lever: a Search generative AI control letting sites limit their content's use in Search's AI responses. Know it exists, and know who it's for. Publishers making a deliberate business-model or licensing stand now have a first-party switch, which is healthier than the robots.txt hacks that preceded it.
For nearly everyone reading a marketing blog, though, the right setting is untouched. Blocking trades presence in the surfaces where shortlists form for a principle most B2B companies don't hold, and the absorption problem it responds to is better answered by becoming the cited source than by leaving the room. If your leadership asks about the toggle after reading a publisher headline, that's the two-sentence answer, and the five tracks above are the evidence you bring.
What To Pair It With
The report's two structural gaps define its two required companions.
No queries means you run the queries. Keep a fixed list of the buying questions that precede your deals, run them in AI Mode and the other engines on a schedule, and log who's cited. The report tells you which pages appear somewhere; your query runs tell you whether you win the questions that matter, and a weekly per-engine tracking loop does it across ChatGPT, Perplexity, Claude, Gemini, and Grok, where Search Console will never see at all.
No clicks means analytics carries the outcome side. AI-surface clicks arrive in analytics as ordinary Google organic, so the pairing is cohort logic rather than a magic segment: page classes, dated changes, and the impressions-versus-sessions divergence read together, with the same discipline GSC's query strings deserve.
One warning shaped by experience with every new Google report: resist rebuilding your reporting around the shiny table for a quarter, then abandoning it. The AI report earns exactly one standing chart in your monthly deck, AI share of impressions with the absorption gap beside it, plus a page-level check when something moves. More than that is dashboard theater; less is flying blind in the fastest-changing surface Google ships.
The Executive Translation
Because this report will be screenshotted into a leadership deck within a month of you opening it, here's the three-sentence script that keeps it honest. "This shows how often our pages appear inside Google's AI answers; it counts appearances, not visits, and Google doesn't tell us which questions triggered them. Our AI share of impressions is X percent and trending this direction, which tells us how much of our visibility now lives inside answers rather than links. The click impact shows up in the gap between these impressions and our session curve, and the question-level view comes from our own citation tracking, which is the other chart on this slide."
Three sentences, two charts, no overclaiming, and it preempts the two questions executives always ask next: "so is this good?" and "which searches are these?"
The One-Paragraph Verdict
Search Console's AI performance report is the first honest window into your presence inside Google's answer layer: real data, page-level, first-party, and finally rolled out to everyone. Read it knowing its counting rules blend presence with reader behavior, track the five lines that translate to decisions, leave the blocking control alone unless you're making a publisher's stand, and pair it with the query-level and cross-engine instruments it structurally can't replace. Exposure is now measurable. Winning the questions is still your job, and still measured elsewhere.
Built by the team that got cited in 48 hours.
Content generation, backlink building, AI visibility tracking, and Google rankings. One platform, zero guesswork.
