Last week, SEOs started noticing something strange in their Google Search Console query reports. Not long-tail phrases. Not misspellings. Just fragments:
yesyes go onyes, pricing
Anastasia Kourou flagged a batch of them publicly and asked John Mueller what she was looking at. His answer, in effect: those are AI Mode. Search Console folds AI Overviews and AI Mode data into the general performance report, and when a user asks a follow-up question inside an AI Mode conversation, Google treats that turn as an entirely new query — with its own impressions, clicks, and position.
So when someone asks AI Mode about dry eye treatments, gets a synthesized answer that cites your page, and then types “yes, pricing” to keep going, that two-word fragment lands in your Search Console like any other search term.
That is a genuinely useful discovery. It is also the only new visibility most site owners have gotten into AI Mode — and it arrives as an unlabeled fragment you have to reverse-engineer.
If you have read a post recently claiming that Google added a dedicated “AI Mode” option to the Search Type filter, or that Search Console now has a natural-language query tool where you can ask it to show your AI Mode traffic, those posts are wrong. Neither feature exists. Let’s walk through what Google actually shipped, what it actually reports, and what you’re still flying blind on.
Where AI Mode data actually lives
Google’s own developer documentation is unambiguous on this point: sites that appear in AI features — both AI Overviews and AI Mode — are counted in your overall search traffic in Search Console, and they are reported inside the Performance report under the “Web” search type.
Under “Web.” Not beside it. Not filterable out of it.
That means:
- There is no AI Mode option in the Search Type dropdown. The search types are Web, Image, Video, News, and Discover.
- There is no Search Appearance filter that isolates AI Overview or AI Mode impressions from classic blue-link impressions.
- Your AI Mode clicks, AI Overview clicks, and traditional organic clicks are one undifferentiated number.
This is not an oversight or a rollout gap. Google has been asked repeatedly for an AI Overviews filter and has declined to provide one. The commingling is the design.
The practical consequence for reporting: when a client’s organic traffic moves, you cannot attribute the move. It might be classic rankings. It might be AI Overview expansion suppressing clicks on queries you still rank #1 for. It might be AI Mode adoption shifting query patterns entirely. It might be seasonality. Search Console will not tell you which, and no amount of filtering will make it tell you.
The report Google did ship — and what it leaves out
There is a new report. It’s called Generative AI features, it sits under Performance, and Google has been rolling it out incrementally since roughly June 2026 — first in the UK, then India, then the US and elsewhere. Not every property has it yet.
Here is what it gives you:
| Dimension | Available |
|---|---|
| Impressions | Yes |
| Pages | Yes |
| Countries | Yes |
| Devices | Yes |
| Dates (hourly through monthly) | Yes |
| Clicks | No |
| Queries | No |
Read that bottom half again, because it inverts the story most coverage is telling. The dedicated AI report has no query data and no click data. You can see that a page of yours surfaced in generative AI features, in a country, on a device, on a date. You cannot see what was asked, and you cannot see whether anyone came.
Meanwhile the main Performance report does have query data — it just won’t tell you which queries came from AI. So the two reports each hold half of what you need, and Google has not joined them.
How the metrics are actually counted
Where the numbers do exist, it’s worth knowing the definitions Google published, because they behave differently from classic search:
Impression. If a link to your page appears in the AI response, that’s an impression. Because AI answers resolve intent on the page itself, broad-topic impressions can climb sharply while clicks stay flat. Rising impressions in the AI era are not automatically good news.
Click. Only an outbound click on an external link in the AI response counts. The upside: users who do click have already read a synthesized answer and clicked anyway. That’s a pre-qualified visitor with a reason to want your specific page — usually depth, proof, pricing, or a transaction.
Position. Calculated the same way as standard search results. Not, as some coverage claims, a measure of where you sat within the AI answer’s visual layout.
Follow-up turns. Each conversational turn is a distinct query with its own metrics. This is the mechanic behind the “yes go on” rows — and it means a single user session can generate several query rows that look meaningless in isolation.
What you can actually do with this today
None of the above makes Search Console useless. It makes it a partial instrument, and partial instruments still work if you know their range. Here’s an honest checklist:
1. Hunt the fragments. Sort your query report by impressions and scan for conversational debris — bare affirmatives (“yes,” “sure,” “go on”), sentence fragments, second-person phrasing, and unusually long natural-language questions. These are your strongest available proxy for AI-driven sessions. Tag them and track the cohort’s size month over month. Growth in that cohort is your AI adoption curve.
2. Check whether you have the Generative AI features report. If it’s live on your property, it will tell you which pages Google is drawing on for AI answers. That page list is genuinely valuable even without queries attached — it tells you what your AI-facing surface area actually is, which is often not what you assumed.
3. Watch the impressions-to-clicks spread. If impressions rise on a page while clicks flatten, you are likely being cited into a zero-click answer. That’s not a failure state — it’s a positioning change. Rewrite the page so the click has a job that the AI summary can’t do: original data, pricing, tooling, comparison depth, primary-source detail.
4. Structure content for conversational retrieval. Question-shaped H2s and H3s with a direct, self-contained answer immediately beneath them. Generative systems extract passages, not pages. A passage that stands alone gets cited; a passage that depends on three paragraphs of preceding context does not.
5. Build entity authority, not keyword density. Generative engines resolve entities and weigh topical consensus. Consistent naming, corroborating third-party mentions, clean structured data that matches visible text, and factual claims that agree across the web all matter more than anything you do at the keyword level.
The ceiling — and it’s a hard one
Now the part that no amount of Search Console skill gets you past.
Search Console is a Google-only instrument. It has zero visibility into ChatGPT, Perplexity, Claude, Microsoft Copilot, Gemini’s standalone app, Grok, DeepSeek, or Mistral. If your buyers research in ChatGPT — and in B2B, many now do — your Search Console dashboard can look perfectly healthy while you are invisible in the surface that actually shapes the shortlist.
It doesn’t measure the things that decide AI outcomes. No competitor share of voice inside a given answer. No sentiment on how your brand is characterized. No unlinked brand mentions, which are frequently how AI systems reference companies. No record of what the model actually said about you — including when it says something false.
That last one deserves emphasis. An AI system can describe your product inaccurately, recommend a competitor over you by name, or cite a three-year-old page as your current positioning. Search Console will register none of it. There is no metric in any Google report for “the model is wrong about you.”
This is the gap CiteMetrix was built to close. It runs your queries against nine AI platforms on a schedule, captures the verbatim responses, tracks citation frequency and share of voice against named competitors, and flags where the answers diverge from your actual brand facts. Search Console tells you about Google’s classic surface. CiteMetrix tells you what every major model is saying about you, in its own words.
You can run a free visibility check on any domain at citemetrix.com/check — no account required. It returns your Free Visibility Score plus the verbatim AI responses, so you can see for yourself which competitors are getting named in your category.
Where to go from here
The measurement layer for generative search is genuinely immature, and it is going to stay that way for a while. Google is not going to hand you clean AI Mode attribution — it has now had three opportunities to and has declined each time. The teams that win the next two years will be the ones that stop waiting for the platform to solve it and stand up their own instrumentation.
If you need help doing that — auditing what Google’s reports can and can’t tell you about your properties, building an AI visibility measurement stack that spans more than one vendor, or training your content and dev teams on how generative retrieval actually works — book a consultation with Expert SEO Consulting. We’ll start with an AI-readiness audit and an honest read on where your blind spots are.
Sources: Google Search Central documentation on AI features and Search Console reporting; Google Search Console Help on impressions, position, and clicks; Search Engine Roundtable reporting on the Generative AI features report rollout and on AI Mode follow-up queries appearing in performance reports (August 2026).
