---
title: ChatGPT query fan-out explained: what is measured and what is not
description: Query fan-out explained: how ChatGPT, Gemini and Google AI Mode split a prompt into searches, what the studies really measured, and how to build your fan-out list.
canonical: https://www.postbing.com/blog/chatgpt-query-fan-out-explained
published: 2026-11-16
updated: 2026-11-16
author: The Postbing Team
---

# ChatGPT query fan-out explained: what is measured and what is not

Query fan-out is when an AI search engine turns one prompt into several shorter searches, reads the results, and writes one answer from them. Google documents it for AI Overviews and AI Mode, and ChatGPT search does it too. The widely quoted "9 to 11 searches per prompt" comes from a study of Gemini 3, not ChatGPT, and no verified per-prompt average for ChatGPT has been published.

Last updated: November 16, 2026

## What query fan-out means

Fan-out is the step between your prompt and the sources. The engine decides it needs more than one search to answer well, generates a set of related queries, runs them, and pools the results. Google's AI features page describes AI Overviews and AI Mode "issuing multiple related searches across subtopics and data sources to develop a response". Fan-out is one of the core ideas in the [generative engine optimization guide](/blog/generative-engine-optimization-guide).

Google's AI optimization guide (May 2026) gives the clearest public example. For the question "how to fix a lawn that's full of weeds", fan-out searches might include "best herbicides for lawns", "remove weeds without chemicals" and "how to prevent weeds in lawn". None of those match the original question word for word, and each could surface a different page.

For you, this means the pages that get cited are the ones that rank for the sub-searches, not necessarily for the prompt.

## How many searches does ChatGPT run per prompt?

Nobody outside OpenAI has published a verified average for ChatGPT, and OpenAI does not document the number. The figures that circulate come from other engines:

| Source | Engine measured | Finding | Sample |
|---|---|---|---|
| Seer Interactive (21 Nov 2025) | Gemini 3 | Average 10.7 fan-out queries per prompt, minimum 3, maximum 28 | 501 prompts via the Gemini 3 API with grounding forced |
| Same Seer study | Gemini 3 vs ChatGPT | "Gemini 3 has 5x the amount of queries that ChatGPT does" | Comparison stated in the study |
| Ahrefs explainer (2 Mar 2026) | Summarizing Seer and Nectiv | "an average of 9-11 fan-out queries per prompt, with 59% triggering 5-11 searches" | Secondary summary |

Two takeaways. First, if you read "ChatGPT runs 9 to 11 searches per prompt", the number belongs to Gemini 3. Second, Seer's own comparison suggests ChatGPT runs fewer searches than Gemini 3, which would make each search ChatGPT does run more decisive. Fan-out also varies by prompt: a simple factual question may trigger one search or none.

## What fan-out queries look like

Fan-out queries are short, generic and repetitive across users. Very different personal prompts collapse into the same few searches. Someone who describes their whole situation in five sentences and someone who types one line can trigger the same "best X for Y" search.

The GEO guide by Nicholas Dulait (November 2026), drawing on a French ChatGPT panel, lists the shapes that dominate:

- best + money keyword + year, sometimes with a country
- the same query in English when the prompt is in another language
- [brand] reviews and [competitor] alternative
- site: searches on authority domains

Here is an illustrative example, written by us rather than captured from ChatGPT. Prompt: "I run a three person bookkeeping firm in Leeds, we hate our current practice software, it's slow and costs too much, what should we switch to?" Plausible searches: "best practice management software for accountants UK 2026", "[current tool] alternatives", "[current tool] pricing", "[tool A] vs [tool B]". The details in the prompt (three people, Leeds, slow) shape how the answer is written, while the searches stay short.

## Why fan-out makes GEO keyword research possible again

Prompts are infinite and unique, so you cannot target them. Fan-out queries converge, so you can. That turns GEO back into keyword research on a known set of query shapes: list the shapes for your category, check which you rank for, and fill the gaps.

It also explains why comparison, alternative and review pages get cited so often for buying questions: those are the shapes the searches take. The page types that match each shape are covered in [six page types AI answers are built from](/blog/six-page-types-ai-answers-are-built-from).

## How to build your fan-out list in an hour

1. **Start from money keywords.** Take the 5 to 10 queries in Search Console that bring you paying customers.
2. **Apply the shapes.** For each, write "best [keyword] [year]", "[your brand] reviews", "[top competitor] alternative", "[you] vs [competitor]", "[competitor] pricing".
3. **Ask a model for sub-questions.** Paste a realistic buyer prompt into ChatGPT or Claude and ask: "List the separate web searches you would need to run to answer this well. Short queries only." Treat the output as ideas, not data. There is a ready set of prompts in [how to use ChatGPT for SEO](/blog/how-to-use-chatgpt-for-seo).
4. **Check what ranks today.** Search each query in an incognito window. Note who ranks and what type of page it is.
5. **Mark your gaps.** Where you have no page, or a page that does not fit the shape, add it to your plan.

Five shapes applied to 5 to 10 money keywords gives 25 to 50 queries before you add the model's ideas. That is a manageable list, not a database.

## What not to do with a fan-out list

Do not create one thin page per fan-out query. Google's AI optimization guide warns that creating pages for query variations "primarily to manipulate rankings or generative AI responses in Google Search" violates its scaled content abuse policy, and that more pages do not make a site more relevant. One strong comparison page can serve "X vs Y", "X or Y" and "X compared to Y" at once.

Also avoid reading too much into a single model's list of sub-questions. The model does not know your market's search volume, and its output changes from run to run. Confirm by checking what actually ranks and, over time, what gets cited in your [weekly citation log](/blog/track-ai-citations-for-free). To turn the list into pages ChatGPT can cite, follow [how to get cited by ChatGPT](/blog/how-to-get-cited-by-chatgpt).

Postbing plans a month of articles from your site's keywords; it does not claim to capture live fan-out queries from ChatGPT. Plans are on the [pricing page](/pricing).

## FAQ

### What is a fan-out query?
It is one of the shorter searches an AI engine generates from your prompt before writing its answer. Google describes AI Overviews and AI Mode issuing "multiple related searches across subtopics and data sources".

### Does ChatGPT run 9 to 11 searches per prompt?
That figure comes from Seer Interactive's study of Gemini 3, which averaged 10.7. Seer said Gemini 3 runs five times as many queries as ChatGPT, and no verified ChatGPT average has been published.

### Can I see the exact fan-out queries ChatGPT uses?
Not reliably. You can ask a model which searches it would run, check which pages it cites, and infer the searches from those pages. Treat model-generated lists as ideas to verify.

### Should I make a separate page for every fan-out query?
No. Group related queries onto one strong page. Google warns that mass-producing pages for query variations to manipulate generative AI answers violates its scaled content abuse policy.

## Sources

- [Google: AI features and your website](https://developers.google.com/search/docs/appearance/ai-features): fan-out definition for AI Overviews and AI Mode, accessed 16 November 2026
- [Google: Optimizing for generative AI search](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide): lawn fan-out example, scaled content warning, accessed 16 November 2026
- Seer Interactive, "Gemini 3 query fan-outs research" (21 November 2025): 10.7 average, 3 to 28 range, 501 prompts, Gemini 3 vs ChatGPT comparison, accessed 16 November 2026
- Ahrefs, "What is query fan-out?" (2 March 2026): 9 to 11 summary and 59% figure, accessed 16 November 2026
- Nicholas Dulait, "The KING of GEO" post on X (13 November 2026): dominant fan-out query shapes from a French ChatGPT panel, accessed 16 November 2026

## Related reading

- [Generative engine optimization: the practical guide](/blog/generative-engine-optimization-guide)
- [Six page types AI answers are built from](/blog/six-page-types-ai-answers-are-built-from)
- [How to get cited by ChatGPT](/blog/how-to-get-cited-by-chatgpt)
- [Postbing plans](/pricing)
