Now live across the AI ecosystem: ChatGPT GPT Store · MCP Registry · mcp.so

Measurement

Your AI Mode questions are already in Search Console. The AI report is not where they live.

By Arnav Mukherjee, founder of TofuBofu · August 12, 2026

At the start of August we removed an engine from our own product. Google AI Overviews had been one of the six surfaces we query on every scan, and when we finally measured what it was contributing, the answer was almost nothing. On buying questions, the kind where somebody asks for the best provider of something, Google returned an AI Overview on 9.1 percent of them. On questions that named a brand directly it answered half the time. The slot we weighted most heavily was the slot that was silent.

So we swapped it for Google AI Mode, which answered the same buying questions three times out of three in the test, with roughly four times the text and a couple of dozen reference links per answer. That fixed our measurement. It also raised a question I did not have a good answer to at the time, which a customer asked me almost immediately: fine, but where does any of this show up in my own analytics?

It turns out the answer is that a lot of it is already sitting in Search Console, for free, and that the report with AI in its name is the one that does not have it.

The report named after AI is the one without the questions

Google has shipped a generative AI performance report in Search Console. Its documentation says it "includes impressions for the following generative AI capabilities on Google Search: AI Overviews, AI Mode", and that Google is "rolling out this report to a subset of website owners, allowing for thorough testing before rolling it further". So not everybody has it yet, and if you do not see it in your property that is expected.

Here is the part worth knowing before you go looking. That report groups impressions by pages, by countries, by dates and by devices. There is no query dimension in it. You can learn which of your URLs got pulled into an AI answer and roughly when, and you cannot learn what anybody asked.

Meanwhile the ordinary Performance report, the one that has been there for years, does have the questions. Google's own documentation for site owners states that "sites appearing in AI features (such as AI Overviews and AI Mode) are included in the overall search traffic in Search Console" and that "they're reported on in the Performance report, within the 'Web' search type".

Read those two facts next to each other and the practical instruction falls out. If you want to know which of your pages AI surfaces are pulling from, use the AI report. If you want to know what people were asking when it happened, do not use the AI report. The queries are downstairs, mixed into the same table as your classic search traffic.

Two Search Console reports, and which one carries the questions

Same AI surfaces, two destinations, one of them has a query column AI Overviews a block above the classic results AI Mode a conversation with follow-up turns Generative AI performance rolling out to a subset of properties pages countries dates, devices no queries Performance report search type: Web classic results AI features, folded in queries, one row per turn Not in either report ChatGPT, Claude, Perplexity, Copilot. Different engines, no impressions feed, no dashboard to subscribe to.

A follow-up question is a new query, and Google says so

This is the sentence that makes the whole thing useful. From Google's Search Console help documentation on impressions, position and clicks: "If a user asks a follow-up question within AI Mode, they are essentially performing a new query. All impression, position, and click data in the new response are counted as coming from this new user query."

The same page confirms the rest of the accounting is ordinary. "Clicking a link to an external page in AI Mode counts as a click." "Position in AI Mode follows the same methodology as a Google Search results page." Nothing exotic. A conversation is being flattened into a stream of queries, one per turn, and dropped into the table you already read every week.

Which explains something people have been posting about for months without a clean explanation: the junk-looking rows. Queries reading "yes", "yes go on", "and pricing", "what about the smaller one". Nobody types those into a search box. They are second and third turns in a conversation, and each one is a moment where your page was still in the room after the opening question had been answered.

That is a genuinely different signal from a head term. A head term tells you somebody found you while assembling a list. A follow-up turn tells you somebody was already deep enough in an answer to interrogate it, and your page was the thing being interrogated. In our own framing, one is discovery and the other is much closer to the shortlist.

What the row looks like
What it probably is
a long, fully specified question
the opening turn, or an ordinary search
a short fragment with no subject
a follow-up turn, subject carried over from before
a bare yes or go on
the user asking the engine to continue
your brand name plus one word
the engine was asked to go deeper on you specifically
a comparison against a rival you do not target
the engine chose your competitive set, not you

A reading guide, not a classifier. Google does not label which rows came from AI Mode, so every line here is an inference you are making.

Search Console shows you one AI surface: Google's own. If you want to know whether ChatGPT, Claude, Perplexity, Gemini, AI Mode and Copilot name you when a buyer asks for a company like yours, that takes a different instrument.

Run a free AI visibility scan

What to do with this on Monday morning

The technique is free, takes about twenty minutes, and needs no tool you do not already have.

Open the Performance report, set the search type to Web, take the longest date range you can, and export the query table. Then stop looking at the top of it. The head terms are the part of your data that behaves the way it always did. Sort ascending by query length instead and read the bottom two hundred rows. That is where the conversational turns concentrate, along with a lot of genuine noise you will have to eyeball past.

Next, pivot those rows onto pages. A page that keeps appearing against fragments is a page an engine is treating as a reference it can return to, which is worth knowing and is not something your rank tracker will ever tell you. Then compare that list against the pages you assumed were your money pages. In my experience they are rarely the same list, and the gap is the interesting part.

Finally, do the writing. For each fragment that reads like a real follow-up, put the answer to that narrower question on the page as its own passage, phrased as the question. This is the same discipline that makes FAQ blocks work for AI answers, and for the same mechanical reason: you are making a specific, self-contained sentence cheap for an engine to lift. SE Ranking's 2026 analysis found 71 percent of ChatGPT-cited pages carry structured data, and 65 percent for Google AI Mode, so if you are adding passages anyway, mark them up while you are in there.

Three things this does not do

I would rather you hear the limits from me than discover them in a board meeting.

It does not separate AI Mode from classic search. Google folds AI features into the Web search type. There is no filter that isolates them, and the report that does isolate them has no queries in it. So you can read the conversational rows and reason about them, and you cannot produce a defensible segment. Telling a short genuine search apart from a conversation turn at scale is unsolved, and anyone selling you a precise split today has inferred it the same way you would.

It does not cover the engines your buyers are mostly using. Search Console is Google's instrument. ChatGPT, Claude, Perplexity and Copilot are outside it entirely, with no impressions feed and no equivalent report. G2's 2026 buyer research puts 51 percent of B2B buyers starting vendor research on an AI chatbot, up from 29 percent. Search Console sees none of that. This is the same blind spot we have written about before, and adding a Google AI report does not close it.

It does not measure conversations, and neither do we. Our own product asks each engine a buying question once per sample and reads the answer. That is a single-turn probe. If a real buyer's decision forms two or three follow-ups deep, then every tool in this category, ours included, is measuring the opening move of a longer game. The Search Console rows are the closest thing to evidence of those later turns that any of us can currently see, which is exactly why they are worth twenty minutes.

None of that makes the data useless. It makes it a floor. SEO has always been the floor, and this is the newest part of it: the one AI surface Google will report on is Google's own, it is free, and most people are looking at the wrong report.

Frequently asked questions

Does Google Search Console show AI Mode data?

Yes, in two different places, and they carry different things. Google's documentation states that sites appearing in AI features such as AI Overviews and AI Mode are included in the overall search traffic in Search Console, and are reported on in the Performance report within the Web search type. Separately there is a generative AI performance report, which Google is rolling out to a subset of website owners. The confusing part is that the report named after AI is the one with less detail in it: it groups impressions by pages, countries, dates and devices. The question-level data is in the ordinary Performance report, mixed in with classic search.

Are AI Mode follow-up questions counted separately in Search Console?

Yes, and Google says so plainly. The Search Console documentation states that if a user asks a follow-up question within AI Mode, they are essentially performing a new query, and that all impression, position and click data in the new response are counted as coming from this new user query. That is why marketers keep finding rows like 'yes' or 'what about pricing' in their query data. Those are not people typing fragments into a search box. They are turns in a conversation, each recorded as its own query.

Why am I seeing strange short queries like 'yes' in Search Console?

Because a conversational turn inside AI Mode is recorded as a new query, and conversational turns are often short, elliptical and dependent on what came before. A row reading 'yes go on' or 'and the pricing' makes no sense as a standalone search and perfect sense as the second or third thing somebody said. Treat these rows as evidence that your page was pulled into an ongoing conversation rather than a single lookup, which is a different and usually later moment in the buying process.

Can I separate AI Mode queries from normal Google queries in Search Console?

Not cleanly, and this is the honest limit of the technique. Google folds AI features into the Web search type rather than giving you a filter that isolates them, and the generative AI report that does isolate them carries no query dimension. So you can see the conversational rows and reason about them, but you cannot produce a clean segment that says these hundred impressions came from AI Mode and those did not. Anyone claiming a precise split today is inferring it, not reading it.

Does Search Console show ChatGPT, Claude or Perplexity data?

No, and it never will. Search Console is Google's instrument and reports on Google's surfaces. The one AI surface it can see is Google's own. ChatGPT, Claude, Perplexity and Copilot are separate engines with no shared dashboard, no impressions feed and no report you can subscribe to. If more than half of B2B buyers now begin vendor research on an AI chatbot, as G2's 2026 buyer behaviour research found, then Search Console covers the discovery surface you already had and none of the ones that arrived since.

What should I actually do with AI Mode query data?

Three things, in order. First, sort your query rows by length and read the short and the odd ones rather than the head terms, because that is where conversational turns concentrate. Second, look at which of your pages those turns land on, since a page pulled into a follow-up is answering a narrower question than the one that started the conversation. Third, write the answer to that narrower question as its own passage on the page, phrased as the question. You are not optimising for a keyword here, you are making a specific sentence easy to lift.

Is AI Mode traffic worth optimising for if the volume is small?

The volume argument cuts both ways and the honest answer depends on which number you look at. Orbit Media's analysis of 97 B2B GA4 accounts covering 28.9 million sessions to June 2026 found AI sources were about 0.5 percent of traffic, which sounds dismissible, and that those visitors converted at roughly three times the rate of other organic sources, which does not. Small and high intent is the normal shape of a channel early. The cheap move is to read the data you already have rather than to fund a programme against it.

Sources and further reading

Related reading