Query fan-out is the mechanism behind AI Mode and AI Overviews, and it is one of the few parts of AI search that Google documents in plain words rather than leaving to speculation. It is also the part people build the most elaborate theories on top of.
So here is what Google actually says, and then what follows from it. Everything in this post that is presented as Google’s position is a direct quote, because on this topic the gap between what is documented and what is asserted is very wide.
What is query fan-out?
From Google’s AI features and your website documentation:
“Both AI Overviews and AI Mode may use a ‘query fan-out’ technique, issuing multiple related searches across subtopics and data sources, to develop a response.”
The AI Mode help page puts it the same way from the user’s side:
“AI Mode groups the user’s question into subtopics and searches for each one simultaneously, and users can go deeper.”
That is the whole mechanism. One question in, several searches out, run at once, results assembled into one answer. It is not a new ranking system. It is several ordinary searches you never see.
Built from Google's descriptions of query fan-out in its AI features documentation and the AI Mode help page.
Does this mean I should optimise for the subtopics?
This is where I part company with most of what is written about fan-out, because Google is unusually blunt here. Same documentation page:
“There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.”
Read that twice. There is no fan-out optimisation checklist from Google, because Google’s stated position is that there is nothing extra to do.
What is genuinely useful about knowing the mechanism is not a tactic. It is a correction to how you read your own data. If one question becomes four searches, then your page did not fail to rank for the question. It either did or did not surface for one of four subtopics you cannot see. Those are different problems and only one of them is solvable.
The practical version of this is unglamorous: a page that answers one thing completely is a better fan-out candidate than a page that answers five things partly, because a subtopic search is narrower than the question that spawned it. That is the same advice as before fan-out existed. It just has a clearer reason now.
Can I see fan-out queries in Search Console?
No, and this is the single most useful thing to know.
AI Mode and AI Overviews performance is reported, per the same documentation: “they’re reported on in the Performance report, within the ‘Web’ search type.” So the clicks and impressions are in there, mixed into your normal Web numbers rather than split out.
The separate generative AI report is more limited than people expect. Its documentation lists the available dimensions as Pages, Countries, Dates and Devices. There is no query dimension. None. You can see which of your pages were shown in a generative AI feature and on what day, and that is it.
Impressions there mean something specific too: “Impressions are how many times links to your site were shown to a user in a generative AI feature on Google Search.” Shown as a link. Not the same as being used to write the answer without a link.
So there is no report anywhere that shows you the subtopic queries. Any tool claiming to show you fan-out queries is inferring them, not reading them. I have written separately about what AI Mode traffic looks like in Search Console and where the numbers do and do not line up.
The follow-up question detail
One line from the AI Mode help page changes how conversational sessions should be thought about:
“If a user asks a follow-up question within AI Mode, they are essentially performing a new query.”
Not a refinement of the first one. A new query, which fans out afresh. So a three turn conversation is three independent fan-outs, potentially a dozen subtopic searches, and your page’s presence in turn one tells you nothing about turn three.
This kills a specific bad idea I have seen floated more than once: writing content designed to sit in the middle of a conversation, as the natural second question. There is no conversational position to occupy. Each turn is a fresh search.
What I would actually do with this
Three things, none of them new, all of them better justified now.
- Write pages that fully answer one question. A subtopic search is narrow. A page that is narrow and complete matches it. A hub page covering ten things matches none of them well.
- Make sure the answer survives extraction. Content that only exists after JavaScript runs is a risk here for the same reason it always was, which I covered in do AI crawlers render JavaScript. Fan-out does not change rendering.
- Measure at page level, because that is the only level available. The generative AI report gives you pages and dates. Use them. Do not try to reconstruct queries you have no access to.
For the wider picture, the AI search visibility guide covers the whole surface rather than this one mechanism, and tracking brand mentions in AI search covers the part fan-out reporting cannot reach.
Sources
- AI features and your website, Google Search Central. The query fan-out definition, the statement that no special optimisations are necessary, and the note that AI features are reported under the Web search type.
- AI Mode, Google Search Console Help. Subtopic grouping and simultaneous search, and the statement that a follow-up is a new query.
- Performance report for generative AI features, Google Search Console Help. The available dimensions, and the definition of an impression in that report.