How to track Google AI Mode traffic in Search Console (2026)
Track Google AI Mode traffic in Search Console: read the generative AI impressions report, isolate AI Mode queries hidden in your standard report, and respond.
Google's AI Mode answers a question inside Search, conversationally, often without sending a click — so the traffic it drives is real but hard to see. As of June 3, 2026, Search Console does report on it: a dedicated Generative AI performance report shows how often your pages appear inside Google's AI features. But it reports impressions only, it folds AI Mode together with AI Overviews, and it hides the queries behind the activity. So "tracking AI Mode" is really two jobs — reading the official visibility report, and reconstructing the query side from data that leaks into your standard report.
This is the workflow for doing both without inventing numbers. The early steps read what Google gives you directly and set the right expectations for it; the later steps isolate the AI Mode queries that surface, disguised as ordinary searches, in the normal Performance report — and scale that from manual scanning to a filter to a full export. Work them in order. The report tells you where you are being cited; the query reconstruction tells you what people are asking to get there; together they are your queue for a content response. For the click-loss side of the same picture, see [how to measure traffic lost to Google AI Overviews](/how-to/measure-ai-overviews-traffic-loss).
The steps
Open the Generative AI performance report first. In Search Console, go to Performance → Search results and look for the Generative AI view. Google rolled it out from June 3, 2026 (staged, so not every property has it yet), with impression data beginning May 18, 2026 and no historical backfill. It reports how often your URLs appeared inside Google's generative AI features — AI Overviews and AI Mode together — broken down by page, country, device, and date. Note what it does not include: no clicks, no CTR, no queries. If you cannot see it, your property may not be in the rollout yet, or the site has not logged enough AI impressions.
Read AI impressions as reach, not as traffic. Because there is no click or CTR column, treat this report as a visibility signal: which of your pages Google is pulling into AI answers, and how often. Sort by page to find your most-cited URLs. The pattern to watch for is a page with high AI impressions whose clicks — over in the standard report — are flat or falling. That is a page AI is answering inline without passing the click along, and that gap is the thing to act on, not a number to celebrate.
Separate AI reach from classic reach in the standard report. Clicks and impressions from AI Overviews and AI Mode are still folded into the aggregate 'Web' search type in the main Performance report, and there is no native filter to peel them out cleanly. Set up a saved date-range comparison instead: baseline your total clicks and impressions from before AI features scaled, then watch the divergence. Impressions holding or rising while clicks soften is the classic AI-answer signature — you are being seen more and clicked less.
Find AI Mode queries hiding in your query report. AI Mode conversations leak into the standard Performance report disguised as ordinary queries. You will see conversational fragments, follow-ups like 'yes go on', 'continue', and 'more', whole pasted prompts, and prompt verbs such as 'write', 'draft', and 'summarize'. Google's John Mueller confirmed this query data had always been in the performance report. Open the Queries view and scan for anything that reads like a spoken sentence or a chatbot reply rather than a keyword — that is your raw AI Mode signal. See [AI conversations in Google Search Console](/guides/ai-conversations-in-google-search-console) for what these strings look like.
Apply a custom regex filter to surface them at scale. Rather than scrolling, use the Query filter's Custom (regex) mode to match those patterns in one pass — greetings, acknowledgements, follow-up cues, prompt verbs, and full question phrasing. It is instant, free, and needs no export. The limit is honesty: an English-pattern regex misses other languages and will not catch every conversational query, so refine the pattern against your own data instead of trusting a copied one wholesale.
Export past the 1,000-row cap for the full picture. The UI caps query tables at 1,000 rows, which buries the long tail where most conversational queries actually live. Pull the complete query set through the Search Analytics API or a GSC export tool, then filter or classify it offline — a spreadsheet rule set, or an LLM or trained classifier that buckets each query as conversational, follow-up, pasted prompt, or ordinary search. Treat every count as a floor: Google anonymizes rare queries, and conversation strings are exactly the rare kind.
Turn the read into a monthly tracking habit. None of this is one-and-done. On a fixed cadence, snapshot three things: AI impressions by page from the Generative AI report, AI-query volume from your regex or export pass, and total clicks versus impressions from the standard report. Watching those lines together over months is how you tell whether your AI visibility is growing and which pages keep earning citations without clicks — which is precisely the queue for the content you produce next. Fold it into your reporting the way you would any other channel; [creating social media reports with AI](/how-to/create-social-media-reports-with-ai) covers the workflow.
Common gotchas
The report shows impressions only — no clicks, CTR, or queries. It tells you that you appeared inside an AI answer, not whether anyone clicked. Always pair it with the standard Performance report to see the click side.
AI Mode and AI Overviews are combined into one bucket. The Generative AI report does not let you split AI Mode impressions from AI Overviews — they are reported together as 'generative AI features'. Any tool promising a clean AI-Mode-only number from native GSC is estimating, not reading.
Regex filters and classifiers undercount. Google anonymizes rare queries, so the conversational strings you can see are a fraction of the real pool — one measurement over 59 days put roughly 58% of impressions in the anonymized pool, leaving only about 42% visible. Read the trend, not the absolute.
The 1,000-row UI cap hides the tail. Conversational queries are long-tail by nature; if you only look at the top rows in the UI, you miss most of them. Export the full query set to see them.
There is no historical backfill. AI-feature impression data starts mid-May 2026, so you cannot see how your AI visibility looked before that — start your baseline now rather than expecting past data.
A staged rollout means absence is not zero. If the Generative AI view is not in your property yet, that is expected and does not mean you have no AI impressions — check back rather than concluding you are invisible.
Where Kompozy fits
This whole workflow ends in a diagnosis, not a fix: a list of pages Google answers inline — high AI impressions, thinning clicks — and a pile of conversational queries proving people reach your topic through AI without ever landing on you. Search Console can show you that gap; it cannot close it. Closing it means publishing more of the specific, liftable content AI answers pull from, on more of the surfaces they pull from, at a cadence your monthly review can actually feed. That production layer is what Kompozy is — a full AI content generation and multi-platform publishing engine, not an analytics tool, and it will not replace the tracking above. From one [Persona Brief](/glossary/persona-brief) it turns each cited-but-clickless page into a coordinated response: a [Blog Article](/glossary/output-buckets) that answers the leaked conversational queries head-on with front-loaded, quotable passages, Carousel Posts and Quote Graphics that restate the load-bearing facts as discrete liftable units, and a [Persona Short](/glossary/persona-shorts) where your named expert says it on camera — video being one of the most-cited sources in AI answers. [Autopilot](/glossary/autopilot) then schedules and fans that set across the eight social platforms plus your blog and newsletter behind a per-post review gate, so the same true claim lands on multiple independent surfaces — the cross-source corroboration AI engines reward. The honest framing: keep the Search Console workflow as your measurement loop and let it point the queue; Kompozy is the engine that clears the queue, turning "we appear in AI answers but earn no click" into more pages worth citing across every channel. It complements the on-page groundwork behind [generative engine optimization](/glossary/generative-engine-optimization) rather than replacing it. Starter ($99/mo, 5,500 credits) fits a solo operator responding to one topic cluster; Pro ($299/mo, 18,000 credits) suits a brand producing across every platform; Enterprise is custom for agencies tracking and feeding AI visibility across clients.
Frequently asked questions
Does Search Console show Google AI Mode traffic separately?
Partly. Since June 3, 2026, a dedicated Generative AI performance report shows impressions from Google's AI features in Search. But it combines AI Mode with AI Overviews into one bucket and reports impressions only — no clicks, no CTR, no queries, and no separate AI-Mode-only figure. For the query side you have to work from the standard Performance report.
Can I see AI Mode clicks in Search Console?
No. There is no AI-specific click or CTR metric. Clicks from AI Overviews and AI Mode are folded into your overall Search performance in the standard report, so you can measure your total clicks but not the AI-driven slice on its own. The Generative AI report deliberately stops at impressions.
How do I find AI Mode queries in Search Console?
They surface in the standard Performance report disguised as ordinary queries — conversational fragments, follow-ups like 'yes go on', pasted prompts, and prompt verbs such as 'write' or 'summarize'. Open the Queries view and filter for those patterns; a custom regex in the Query filter catches them in one pass, or you can export the full query set and classify it offline.
Why does the Generative AI report not show queries?
Google exposes impressions, pages, countries, devices, and dates in that report, but not the queries behind them, and it is UI-only — not available through the Search Analytics API or the BigQuery bulk export. The query signal lives in the standard Performance report instead, where AI Mode conversations appear as ordinary-looking search queries.
When did AI Mode data arrive in Search Console?
Google announced the Search Generative AI performance reports on June 3, 2026, rolling them out to a subset of properties first. Impression data starts from May 18, 2026, with no historical backfill, so the earliest AI-visibility baseline you can build begins around that date.
Is the AI Mode query count from Search Console accurate?
Treat it as a floor, not a total. Google anonymizes rare queries, and conversation strings are rare, so a meaningful share never appears in your data — a measured undercount has left a large fraction of impressions in the anonymized pool. Use the numbers for direction and trend, not for precision.