Local query fan-outs: the hidden searches behind "best X in Y"

A prompt like "best orthodontist in Rotterdam" looks like one search. When an AI engine answers it with live search, it usually isn't. The engine rewrites the question into narrower searches, reads the results, and writes one answer from them. Those narrower searches are query fan-outs, and we cover the general idea in query fan-outs explained.

Local prompts add a twist. Location gets carried into the fan-outs, and sometimes it gets added even when the person never typed one. That changes which pages compete.

What the vendors document

OpenAI's help page on ChatGPT search says that when ChatGPT uses a search partner, it "typically rewrites your query into one or more targeted queries." It also says ChatGPT collects general location from your IP address and may share that general location with search providers. OpenAI's own example is the clearest description of a local fan-out we've seen: someone in the San Francisco area asks for "good restaurants near me", and ChatGPT may rewrite it into the search "top restaurants San Francisco." With memory on, a saved detail can go in too. OpenAI's example there is "good vegan restaurants San Francisco" for someone who has said they're vegan and live in the city.

Google's documentation on AI features says AI Overviews and AI Mode may use a query fan-out technique, "issuing multiple related searches across subtopics and data sources." It doesn't describe how location enters those searches.

That's as far as the documentation goes. What follows is a way to read the fan-outs once you collect them, not a description of how any engine works inside.

What to look for in local fan-outs

When you line up the fan-outs behind a set of city prompts, check how the place appears in each one.

Is the city from the prompt carried into the narrower searches, or dropped from some of them? A fan-out that loses the city is a national search, and national pages will compete for it.

Was a location inserted where the prompt had none? That's the documented ChatGPT case: "near me" becomes a named city based on IP. It's also why an implicit prompt like "emergency dentist" is worth running from more than one place and comparing.

Did the location get narrower or broader? A city prompt may produce searches about a district, and a district prompt may widen to the whole city. Either tells you which level of page the engine went looking for.

To make that concrete, here's an illustrative example. It wasn't captured from a real run. A prompt like "best orthodontist in Rotterdam for adults" might fan out into searches along the lines of:

adult orthodontist Rotterdam
invisible braces cost Rotterdam
orthodontist reviews Rotterdam
a comparison between two practices the engine already associates with the city

Your real fan-outs will look different. What matters is the shape: one local question becomes several local subtopics, each a separate chance to be the best result.

Why it matters for local businesses

The pages cited in the final answer come from the results of those narrower searches. A practice with a weak homepage but a clear, specific page on adult treatment costs can win the "cost" fan-out and get linked, even if a bigger competitor owns the head query. And if every fan-out for your category carries "reviews", the answer will lean on review sources, which no page on your own site can stand in for.

Seeing the fan-outs for yourself

Start with your own browser. Promptwatch's free GEO Inspector Chrome extension shows the fan-outs ChatGPT ran for a conversation and which sources it fetched versus which it actually cited. It runs locally. Keep in mind that it shows what ChatGPT did for you, from your location, so a test from head office reflects head office.

That fetched-versus-cited split is useful on local prompts. If your location page was fetched but not cited, the engine found you and preferred something else, which is a content question. If it never appears in the fetched list, look at access and at whether your page would rank for the fan-out at all.

Turning fan-outs into a local content plan

The process is the same as for any fan-out work, with one extra dimension.

Collect fan-outs for each city prompt over several runs and more than one engine. One run tells you very little.
Cluster them by subtopic within each city: cost, availability, specific services, reviews, named comparisons.
Count which clusters recur across cities. A "same-day appointments" cluster under every city is a template change for all location pages. One that appears in a single city is a local page edit.
Map each cluster to a page that should answer it, and check whether that page gets fetched and cited afterwards.

One warning. Fan-outs that name districts or suburbs aren't an invitation to generate a thin page for every neighborhood. Google's spam policies list "having multiple domain names or pages targeted at specific regions or cities that funnel users to one page" as an example of doorway abuse. Write a page for a place when you actually serve it and have something specific to say about it.

Tracking fan-outs from the right location

Doing this by hand from one browser doesn't scale, and it shows only your own location. For client work we use prompt tracking in Promptwatch, which records the query fan-outs per model behind each tracked prompt. Put the prompts in a monitor set to the city (state and city tracking starts on Business and is on every agency plan), or pin it to exact coordinates with local GEO/SEO targeting on Business, Agency Scale, and Enterprise. With coordinates set, ChatGPT, Google AI Overviews, and Google AI Mode answer as if asked from that spot, so the fan-outs you review belong to a searcher there and not to your office. Citation analytics sits next to them, so when a page written for a cluster starts getting cited, you see it on the same prompt.

If you'd like us to map the local fan-outs for your locations, write to hello@visiblyrank.com.