Field Guide
Query fan-out: what ChatGPT searches before it answers
When an AI assistant looks something up before answering, it rarely searches for the question it was asked. It breaks the question into several narrower searches, runs them, and writes its answer from what comes back. That step is called query fan-out, and the searches it produces decide which pages the assistant reads. If you are not on those pages, you are not in the answer.
What query fan-out is
A buyer’s question is usually too broad to search for directly. “Is this tool a good choice for us?” has no single page that answers it. So the assistant fans it out into the pieces that do have pages: what it costs, what people say about it, what the alternatives are. Google described this technique publicly when it launched AI Mode, and the assistants that browse, such as ChatGPT with search, Perplexity and Gemini, work the same way: they rewrite one question into a handful of searches before reading anything.
What the searches look like
These come from our own query fan-out model, the one behind the likely searches in meraGEO. They are predictions of what an assistant would search, shown exactly as the model produced them.
“Is PostHog a good choice for developer-first product analytics?”
- PostHog pricing features
- PostHog reviews Reddit G2 Capterra
- PostHog review pros cons 2026
“What is the best invoicing tool for freelancers in the UK?”
- best invoicing software for freelancers 2026
- Tally invoicing free plan
- Wave invoicing features pricing
- "QuickBooks Online" invoicing features
Four patterns worth noticing
Pricing gets its own search. Almost every fan-out includes one. If your pricing is hidden behind “contact sales” or buried in a PDF, the assistant answers that part of the question from someone else’s page, often a competitor’s comparison.
Reviews are searched by name. “Reddit G2 Capterra” is spelled out in the search. The assistant goes looking for third-party opinion specifically, so a thin or missing review profile is a gap it will notice.
The year is in the query. “2026” appears again and again. Roundups and comparison pages that carry the current year are what those searches find, and an undated page from 2023 loses to them.
Competitors are searched directly. Ask about the best invoicing tool and the fan-out names specific products. If you are not already one of the names the model reaches for, you need to be on the pages that list several, such as “best of” roundups and alternatives pages.
What to do with them
Treat each search as a page that should exist and should mention you. For your own site, that usually means a pricing page that states plans and prices in plain text, a comparison or alternatives page for each competitor that keeps coming up, and an FAQ that answers the narrow questions in the buyer’s words. Off your site, it means review profiles that are claimed and current, and a place in the roundups that already rank for the “best X 2026” searches. The playbook for showing up in AI answers covers each of these in more depth.
Remembering versus searching
An assistant answers in one of two ways. Without search, it answers from what it learned in training, and fan-out plays no part. With search, it answers from the pages its fan-out finds. A meraGEO scan measures the first: it asks ChatGPT, Claude and Gemini without web search, which shows what the models already believe about you. The likely searches cover the second, and they are predictions, not searches we watched an assistant run. The two together tell you what a model thinks of you today and where it would look to check.
See the likely searches for your brand
Run a free scan, then sign in: each question we asked shows the searches an assistant would likely run for it. On a paid plan you can add the questions your buyers actually ask and get the same for those, every week.
Run a free scan