On 3 June 2026, Search Console got Search Generative AI performance reports, two of them in fact: one covering AI Overviews and AI Mode in Search, and a separate one for generative features in Discover. Impressions from the answer layer, broken out for the first time. It is genuinely useful. It is also impressions-only. No clicks. No click-through rate. No query data in the report. Every link inside an AI Overview shares a single position value, so you can't even tell whether you were cited first or fifth.
So you can now prove you're visible in AI answers. You still can't prove it's worth anything. This is how to close that gap with data you already have.
What the report actually gives you
Worth being precise about the scope, because a lot of the commentary got it wrong.
| Dimension | Available? | Notes |
|---|---|---|
| Impressions | Yes | Across AI Overviews, AI Mode, and Discover AI features |
| Pages | Yes | Which of your URLs surfaced |
| Countries | Yes | Useful given the staggered rollout |
| Devices | Yes | Search surfaces only |
| Date | Yes | Hourly through monthly granularity |
| Clicks | No | Not reported at all |
| CTR | No | Follows from the above |
| Queries | No | Not in this report, see note below |
| Position | Partial | One shared value per AI response, not per link |
Three pieces of context that change how you read it.
These impressions were never missing. Google confirmed generative AI impression data was already inside your overall performance report, "where it will continue to be tracked". The new view is a dedicated cut, not additional volume. If you were about to email a client explaining a sudden traffic increase, don't. Your totals haven't moved.
Queries aren't gone, they're just elsewhere. John Mueller confirmed on 6 August 2026 that AI Mode queries do appear in the main Search Console performance report. They're absent from the generative AI report specifically. So the keyword-level view isn't a total blackout, it's unsegmented, which is a different and slightly more workable problem.
Also worth knowing how an impression is counted. Google's documentation specifies that a link inside an AI Overview must be scrolled or expanded into view to register. Collapsed citations nobody ever revealed don't count, which makes the number more meaningful than a raw render count would be.
On rollout: a subset of UK site owners got early access, alongside a toggle that blocks your content from appearing in AI features. The toggle is the piece the Competition and Markets Authority actually ordered on 3 June, as part of its publisher controls and attribution requirements; the reports arriving at the same moment is likely related but wasn't itself mandated. Access has since been widening beyond the UK. Two details on the toggle: it is explicitly not used as a ranking signal for search outside the generative features, and it does not cover the Gemini app.
Why impressions alone will mislead you
An impression in an AI Overview and an impression in a blue link are not the same event, and averaging them together produces a number that means nothing.
A classic result appears in a list the user is actively scanning for something to click. An AI citation appears as a footnote next to a synthesised answer that may have already resolved the query. The intent to click has, in many cases, been consumed before your citation is even read.
Which means an AI impression count that goes up is ambiguous on its own. It could mean growing authority in the answer layer. It could equally mean you're being harvested more efficiently. Impressions tell you the direction of visibility. They tell you nothing about the direction of value.
Filling the gap with GA4
You can reconstruct a workable proxy for the missing half, because the referral side was solved a few weeks before the Search Console side.
Since around 13 May 2026, GA4 has had an automatic AI Assistants channel. Google's definition names ChatGPT, Gemini, DeepSeek, Copilot and Grok as example sources, and explicitly excludes Google's own AI Overviews and AI Mode. That is your click side. If the channel isn't reporting cleanly yet, our guide on how to track AI traffic in GA4 covers the property setup this method assumes.
The method:
- Pull AI impressions by page from the Search Console gen AI report for a fixed window. Monthly is the right granularity to start; anything shorter is too noisy at low volume.
- Pull AI Assistants sessions by landing page in GA4 for the same window, same date boundaries.
- Join on URL. You now have an approximate AI CTR per page.
- Segment by page type. Comparison pages, definitional pages, and how-to guides behave very differently. A blended site-wide figure hides the only insight worth having.
Treat the output as directional, not exact, and understand precisely why. Google's report counts impressions in Google's AI surfaces. GA4's AI Assistants channel counts referrals from third-party assistants, with Google's own AI features deliberately excluded. These are not two halves of one funnel. What you're building is a rough visibility-to-value ratio across the answer layer as a whole, not a true CTR for either surface.
That's still worth having. It just isn't the number Google withheld, and you shouldn't present it as though it were.
The Perplexity hole
One omission to patch: Perplexity is not among the sources Google names in the AI Assistants channel definition, and in practice its referrals still land in Referral or Direct. Google hasn't published a complete referrer list, so treat this as observed behaviour rather than a documented exclusion, and check your own data before assuming you're covered.
Given Perplexity cites sources on essentially every claim and runs its own crawler, it is frequently one of the higher-converting assistant referrers for smaller sites. Leaving it uncounted will systematically understate your AI channel.
Fix it with a custom channel group in GA4 that captures perplexity.ai in the source dimension and folds it in alongside the defaults. Ten minutes of admin work. Do it before you build any reporting on top of the channel, because GA4 channel groups apply going forward and retroactive reprocessing is limited.
Add Bing's citation data
If you want a third reference point, Bing Webmaster Tools gained AI visibility insights on 16 June 2026: Intents, Topics, Citation Share and Compare in the AI Performance dashboard. Citation Share is the closest thing any engine has offered to a first-party "share of AI answer" metric, and Compare puts it against competitors.
This is worth having in its own right, for Copilot visibility if nothing else. A caveat on the folklore, though: the widely repeated claim that ChatGPT's browsing runs on Bing's index is no longer safe to state as fact. Independent tests through 2025 and 2026 have found ChatGPT citing pages that appear in Google's index and not Bing's, and OpenAI operates its own crawler layer. Nobody outside OpenAI can currently say what the retrieval stack looks like. Use Bing's data as a second reading on the answer layer, not as a proxy for ChatGPT.
What to do with the numbers
Once you have impressions and an approximate CTR joined by page, the useful patterns fall out quickly:
- High impressions, near-zero referrals. You are being used as a source, not a destination. The answer is complete without you, and the randomised evidence on AI Overview click loss suggests the visits you're not getting were not junk visits. Either the page needs a reason to click through that the summary cannot replicate (a tool, a calculator, proprietary data, a template), or you accept it as brand exposure and stop measuring it as traffic. On the format side, what actually gets a small site cited is a different question from what ranks, and worth treating separately.
- Low impressions, high referrals. Rare and valuable. You are cited infrequently but persuasively. Find out what those pages do differently and replicate the pattern.
- Both rising. The page is working. Note the format and reuse it.
- Both falling while classic Search holds. Check technical access before you touch content. Blanket bot rules in robots.txt, or a Cloudflare bot category set too aggressively, will quietly remove you from the answer layer while leaving your rankings untouched. Check what you are actually permitting before you touch anything else, and note that llms.txt does not grant access on its own. Note the date on this one: from 15 September 2026, Cloudflare's defaults change on ad-monetised pages, with Training and Agent blocked by default and Search allowed. If you're on Cloudflare and monetising with ads, that switch will flip for you whether or not you've thought about it.
The point
The Search Console AI report is a visibility report wearing a performance report's clothes. On its own it will tell you a flattering story about being everywhere and let you avoid asking whether being everywhere is doing anything.
Join it to GA4, patch the Perplexity gap, and check it against Bing's Citation Share. That gives you visibility and value in the same view, which is the minimum needed before you commit any real budget to AI search optimisation.
The sites that get this right will spend the next year optimising against measured outcomes. The ones that don't will spend it optimising against an impression count that only ever goes up.
Sources
- Google Search Central, "Introducing Search Generative AI performance reports in Search Console," 3 June 2026
- Google Search Console Help, performance report documentation (AI Overview position and impression counting)
- Search Engine Land, "Google Search Console AI performance reports and controls to block your content in AI responses"
- Search Engine Roundtable, on AI Mode queries appearing in the main performance report, 6 August 2026
- CMA, "CMA secures fairer deal for publishers and improves Google Search services in UK," 3 June 2026
- Google Analytics Help, AI Assistants default channel definition, May 2026
- Bing Webmaster Blog, "New AI Visibility Insights in Bing Webmaster Tools: Intents, Topics, Citation Share, Compare," June 2026
- Cloudflare, AI traffic options and September 2026 default changes