Query Fan-Out: What AI Mode Actually Searches When Someone Asks for a London Provider

When someone types 'best dentist near Bank' into a normal search box, one query goes in and ten blue links come out. When the same person asks an AI feature, something different happens behind the scenes, and almost nobody running a local business knows about it.

Google has now documented that mechanism in plain language. It matters because it changes what 'being found' means, and because the obvious reaction to it, writing a page for every sub-query, is exactly the thing Google tells you not to do.

What is query fan-out?

Query fan-out is a set of concurrent, related queries that the model generates to request more information and fetch additional relevant search results for a user's question. That is Google's own definition in its guide to generative AI features. Both AI Overviews and AI Mode may use it, issuing multiple related searches across subtopics and data sources to build a response.

Google's worked example is not local. For the question 'how to fix a lawn that's full of weeds', it says fan-out queries might include 'best herbicides for lawns', 'remove weeds without chemicals' and 'how to prevent weeds in lawn'. One question becomes several, each answered from different pages, and the response draws on the lot.

What is the difference between AI Overviews and AI Mode?

AI Overviews appear on results pages when Google's systems decide they add something beyond classic Search, and AI Mode is a conversational experience a person chooses to use for more complex questions. Google says AI Overviews often do not trigger at all, and that AI Mode is particularly helpful where further exploration, reasoning or comparison is needed.

Scale is changing quickly. At its 19 May 2026 Search announcement, Google said AI Mode had passed one billion monthly users a year after launch, with queries more than doubling every quarter, and that Gemini 3.5 Flash had become the default model in AI Mode globally. I would not build a strategy on a single usage figure, but the direction is plain: more of the research stage of a purchase now happens inside these features.

Both may use different models and techniques, which Google says means the responses and the links they show will vary. That variation is why chasing a particular AI answer is a poor use of time.

What might fan out when someone asks for a London provider?

A request for a local provider plausibly breaks into sub-questions about location, price, trust, comparison and process. The following is an illustration of the mechanism using Google's own pattern, not a record of real Google queries, because Google does not publish the sub-queries it generates.

Notice what those sub-queries have in common. None of them is the head term. They are the practical questions a thoughtful buyer would ask next, and a site that answers them clearly gives the system more to retrieve from.

Does this mean I should write a page for every sub-query?

No, and Google says doing so primarily to manipulate rankings or AI responses violates its scaled content abuse spam policy. The guide is explicit: while it might be tempting to create separate content for every variation, including fan-out queries, a high quantity of pages does not make a site higher quality or more relevant.

That sits awkwardly with a lot of AEO advice I see, which amounts to 'map every possible prompt and publish a page for each'. Google has said in writing that this is both against policy and ineffective. I explain the practical limits of programmatic pages in the post on borough pages.

How do I cover the sub-questions without spinning up pages?

1. List the follow-up questions a buyer asks after the first one. Use your phone log, your enquiry emails and your Search Console queries rather than guesswork.
2. Answer each one inside the relevant page with a clear heading and a direct first sentence. Google says clear headings and organised sections help readers, and a heading is a perfectly good unit.
3. Reserve separate pages for topics with their own intent and enough to say: a distinct service, a distinct audience, a distinct place you genuinely serve.
4. Keep facts consistent across the page, the profile and any third-party listing. Retrieval pulls from several sources at once, and a contradiction between them weakens confidence in all.
5. Add first-hand detail to each answer. That is what separates a page that gets cited from one that gets summarised away.

Why does the Business Profile matter in a fan-out world?

Because several of the sub-queries are answerable only from the profile: opening hours, address, phone number, services offered and reviews. Google says AI responses can include information about local businesses and that Google Business Profiles help your business be visible in them.

Query fan-out turns a thin profile into a bottleneck. If the question 'is it open on Saturday' is part of the sub-search and your hours are wrong, the answer comes from your profile or from a competitor's. The same goes for services you offer that you never listed. I treat the profile as a first-class source for every local AEO piece of work, and I explain the ranking side in the street-by-street post.

How does fan-out change keyword research?

It shifts the unit of research from the keyword to the question cluster. The old habit was to find a head term, list its variants and write something for each. The more useful habit now is to ask what a buyer wants to know between 'I need a provider' and 'I have booked one', and make sure those answers exist and agree with each other.

Google is explicit that you do not need to capture every variation. Its guide says the systems can understand synonyms and general meanings of what someone is seeking, which means you do not have to worry that you lack enough long-tail keywords or have not covered every variation of how someone might ask. That removes a whole category of busywork.

• Start with real questions. Phone calls, enquiry forms, chat logs and the questions customers ask at the end of a job are better than any tool's suggestions.
• Group them by stage. Choosing, comparing, price, process and trust each need a clear answer somewhere on the site.
• Look at your Search Console queries, not for volume, but for phrasing. They show how real people word the same need.
• Check the competitors' pages for gaps, not for ideas to copy. The question nobody answers well is the opportunity.
• Write the answer once, in the right place, and link to it from where the question naturally arises.

The output is a shorter list of better pages and a set of clear headings inside them. That is also the format that is easiest for a person to scan, which is the test Google keeps returning to.

How should I measure whether it is working?

Use the Generative AI performance report in Search Console, and treat third-party prompt checkers as spot checks. Google points site owners to that report for visibility in generative AI features. Prompt-sampling tools record what an AI returned on one occasion, which is a snapshot of a variable system.

The metric I care about is still enquiries. Count calls, form fills and booked jobs, and ask new customers how they found you. For a local business a handful of well-qualified customers who arrived through an AI answer is worth more than a graph of impressions, and Google itself notes that clicks from AI Overviews tend to be of higher quality.

What would I do first?

• Ask the three follow-up questions your best customers asked before they booked, and check your service page answers them.
• Verify the profile against reality: hours, categories, services, phone, photos.
• Stop commissioning one page per prompt and consolidate anything duplicated.
• Add one piece of first-hand evidence to each key page: a real job, a real number, a real name.

Fan-out is not a new thing to optimise for. It is a more demanding reader of the same work. A business that answers real questions clearly, with accurate facts, in more than one place is already doing what it asks for.