Google Finally Shows Your AI Search Data. It Left Out the Two Numbers That Matter.

For two years, the single most requested feature in the SEO world was one number: how much traffic is Google’s AI actually sending me? On June 3, 2026, Google answered, sort of. It launched Search Generative AI performance reports in Search Console, a dedicated view of how your pages show up inside AI Overviews and AI Mode.

Then you open it. It shows impressions. Only impressions. No clicks, no queries. For the two questions every marketer actually has — did anyone come, and what did they ask — Google shipped a blank.

That’s not a gap the rollout will quietly fill next week. It’s the design. And once you see why, you’ll read this report very differently.

What the report actually contains

The new report lives under Performance, as a separate view beneath your Search Results report. You can jump to it by adding /ai to your performance URL. It’s rolling out to UK sites first, with the related AI control settings taking effect June 17, 2026, per Brodie Clark’s early walkthrough.

The honest inventory of what you get and what you don’t:

What the report shows	What it leaves out
Impressions — times a URL from your site appeared in an AI feature	Clicks
Pages that appeared	Queries and prompts
Countries	Position, or which citation slot you filled
Devices (Search only)	Whether you were quoted or just surfaced
Daily, weekly, monthly, back to May 18, 2026	Anything outside Google — ChatGPT, Perplexity, Gemini, Copilot

It covers AI Overviews, AI Mode, and AI Overviews inside Discover. That’s a real expansion of what Google was willing to show even a month ago. But notice what’s missing from the left column. The two metrics you’d use to judge whether any of this is working, clicks and the queries behind them, aren’t there. Clark’s read after seeing it live: the absent metrics are “the most important metrics to assess the impact of AI-related features,” which leaves the current data “nowhere near actionable enough to be used in a practical sense.”

Impressions tell you that you appeared. That’s a weaker claim than it sounds.

An impression here means one thing: a URL from your site showed up somewhere in an AI feature for some query. It does not mean Google quoted you. It does not mean you were the source that shaped the answer. And it definitely doesn’t mean anyone read your name, let alone clicked.

That distinction is the whole game in AI search, and it’s exactly where teams fool themselves. We’ve written before about the gap between getting cited and getting your brand named. Kevin Indig’s data showed that 61.7% of AI citations never say the brand’s name at all. An impression count sits one rung below even that. It confirms your link was eligible to be seen. Whether it did any work for you is a question this report cannot answer.

So a rising impression line feels like progress and might be nothing. You could appear in ten thousand AI Overviews as the fourth link nobody scrolls to, and the chart climbs the same as if you were the pull-quote at the top. Google also doesn’t tell you which of its five separate AI citation slots you landed in, and those slots don’t reward the same content or send the same traffic. One number is standing in for a dozen very different outcomes.

Impression counts have another problem: they’ve been shaky. Clark notes impressions “have been quite unreliable in recent times.” Building a KPI on a metric that’s both incomplete and noisy is how you end up confidently wrong.

The clicks were never going to show up

Now the part that should end the “they’ll add clicks later” optimism, at least for AI Mode.

Links inside AI Mode carry a noreferrer attribute. That attribute strips the referrer value a browser normally passes when someone clicks through. Without it, the visit lands in your analytics as direct or unknown traffic, and Search Console records no click at all. As Search Engine Land documented, this isn’t a reporting delay — it’s baked into how the links are built. AI Mode traffic is structurally hard to attribute.

Lily Ray put the uncomfortable interpretation on the record: “Google does NOT want us having access to traffic data for AI Mode, or AI Overviews for that matter, because it will reveal just how little traffic both are actually driving to external websites.”

Read cynically or charitably, the practical result is the same. When Google eventually bolts click data onto this report, AI Mode clicks may stay invisible anyway, because the plumbing that would carry them was removed on purpose. You’re not waiting for a feature. You’re waiting for a decision Google has quietly already made.

This is the same attribution hole showing up in a new place. It’s why AI-referred traffic keeps getting misfiled as “Direct” in analytics, and why so many teams are running on numbers that miss the fastest-growing part of their referral mix.

The bigger blind spot is the one nobody’s naming

Suppose Google fixed all of it tomorrow: clean clicks, full queries, reliable impressions. You’d still have a report that only sees Google.

That matters more than it seems, because AI answers are not a Google monopoly, and the platforms barely agree with each other. Our own read of the cross-engine data found that ChatGPT and Perplexity cite the same domains only 11% of the time, and Google’s own two AI surfaces overlap on just 13.7% of cited URLs. A win in AI Overviews predicts almost nothing about where you stand in ChatGPT, Perplexity, or Gemini. Each engine reads a different slice of the web and rewards different content.

So a maxed-out Search Console AI report would still be a mirror that reflects one room in a house with five rooms. If your reporting treats “Google’s AI view” as “our AI visibility,” you’re doing the thing we keep warning against: inferring a whole ecosystem from one sample. The number will look authoritative because it comes from Google. It’s still partial by construction.

What it’s genuinely good for

None of this makes the report useless. It makes it narrow, which is different. Used for what it can actually do, it earns a place in the stack:

A first-party sanity check on Google specifically. No sampling, no third-party estimate. This is Google telling you which of your pages it surfaced in its own AI features. That’s a clean signal you couldn’t get before.
Page-level diagnosis. If a page you expected to appear shows zero AI impressions over weeks, that’s a content or structure problem worth chasing, not noise.
Directional trend on Google’s surfaces. Watch the shape of the line over months, not the daily reading. Rising impressions across a topic cluster suggest Google is pulling you into more AI answers, even if you can’t yet prove what that’s worth.
Geographic reality checks. The country breakdown can expose that your AI presence is concentrated in markets you don’t sell to, or missing from ones you do.

Treat it as one instrument on the dashboard. Don’t let it become the dashboard.

How to read it without fooling yourself

The trap isn’t the report. It’s promoting an impression count to a headline metric because it finally exists and it comes from Google. Keeping it in its lane takes a few rules.

Don’t do this	Do this instead
Report “AI impressions” as your AI visibility number	Report it as “Google AI surface appearances,” one line among several engines
Celebrate a rising impression line	Pair it with clicks and conversions from your own analytics before claiming impact
Assume an impression means you were cited or quoted	Treat it as eligibility to be seen, and verify actual citations separately
Infer ChatGPT or Perplexity standing from Google’s data	Track each engine on its own, because they barely overlap
Panic over a daily dip	Judge the multi-week trend on stable pages

Three moves make the report pull its weight:

Bolt it onto your own attribution, don’t replace it. Google’s impressions plus your server logs, analytics, and a real citation tracker across engines gets you closer to truth than any single source. The report is an input, not an answer.
Separate “did we appear” from “did it work.” Use AI impressions to confirm presence. Use clicks, assisted conversions, and branded-search lift to judge value. Keep the two questions apart, because Google is only answering the first one, and only for itself.
Cover the other four rooms. If your AI visibility program starts and stops inside Search Console, most of the AI-answer web is a blind spot. Watch ChatGPT, Perplexity, Gemini, and Copilot directly, because the metrics that actually move competitive visibility live across all of them, not in one vendor’s console.

The takeaway

Give Google credit for the direction. Two years of “we can’t show you that” turned into a real, dedicated report, and the AI controls arriving June 17 suggest more is coming. This is progress.

But progress isn’t the same as enough. What shipped tells you that you appeared, on Google’s surfaces, sometimes, according to a metric that’s been unreliable. It won’t tell you whether anyone clicked, what they asked, or where you stand on the four platforms Google doesn’t own. Those are the numbers that decide whether AI search is sending you customers or just eligibility.

So use the report, for exactly what it measures and not one inch more. The teams that get burned in the next year will be the ones who saw a green line in Search Console and called it visibility. The teams that win will be the ones who kept asking the harder question Google still won’t answer: across every engine, did it actually work?

RivalHound tracks your brand’s visibility across ChatGPT, Google AI, Perplexity, and more. Start monitoring to see where you stand.