Query Fanouts: How ChatGPT Searches the Web
Klaas FoppenWhat Are Query Fanouts?
A query fanout is when ChatGPT breaks a single user prompt into multiple web searches. Instead of running one search, it generates 3-8+ targeted queries, each exploring a different angle of the question, then combines the results into a single answer.
Average Fanouts Per Response
The average number of web search queries ChatGPT generates per response. Higher fanout means ChatGPT is casting a wider net to gather information before answering.
What this means for you
Through the winter, every ChatGPT prompt triggered about two distinct web searches on average: 2.15 in early December, easing to roughly 1.84 by early March. Your content didn't need to match the user's prompt; it needed to match one of the queries ChatGPT generated from it. After a gap in the data, fanouts returned in April at exactly 1.0 per response, a markedly leaner search pattern than before.
How to act on it
- Cover the sub-questions a fanout splits into: comparisons ("X vs Y"), pricing, alternatives, and how-tos. Each fanout query is a separate retrieval your content can win, even when the user never typed those words.
- Favor a cluster of focused pages over one mega-page. Two distinct queries per prompt means two distinct intents being searched, and a page laser-targeted at one intent beats a page that half-covers both.
- Note the direction of travel: from 2.15 queries per response in December down to a single query by April. Fewer searches per answer means each retrieval carries more weight. It's the same slot squeeze happening with citations.
Average Character Count Per Query
The average number of characters in each fanout query per day. Longer queries may indicate more specific or complex search behavior.
What this means for you
ChatGPT's search queries have gotten dramatically more terse: from about 117 characters in early December to the high 80s through February and March, then roughly 53 characters once fanouts returned in April. That's less than half their original length. ChatGPT is searching more like a person typing keywords into Google and less like someone pasting in a full sentence, which changes which headings and titles its queries will match.
How to act on it
- Write headings that read like search queries, not conversational questions. "Best CRM for small agencies 2026" matches a ~53-character query far better than "What should you look for when choosing a CRM?"
- Front-load the entity and category in titles and H2s. At six to eight words per query, terms buried at the end of a long heading contribute little to the match.
- Re-audit your keyword-to-heading mapping after major ChatGPT updates. The drop from 117 to 53 characters happened in steps tied to behavior changes, so phrasing that matched last quarter's queries may already be stale.
How we collect this data
We collect millions of prompt responses, citations, and click data from the actual user interfaces of major AI platforms: over 26 billion data points and growing. This gives us one of the largest datasets on how AI search engines cite sources and recommend brands.
Real UI monitoring
Data straight from the interfaces of ChatGPT, Gemini, Perplexity, Claude, AI Overviews, and more.
26B+ data points
Over 26 billion analyzed citations, prompts, and responses, one of the largest AI search datasets available.
Continuously updated
Refreshed constantly so the trends you see reflect the latest behavior of AI search engines.
Aggregated & public
Published freely for the GEO community, based on aggregated, non-identifiable trends.
Want to start tracking your own AI search data? Get started with Promptwatch
Track Your Brand's AI Search Visibility
Understand how AI search engines break down your topics into sub-queries. Monitor query fanout patterns to optimize your content for ChatGPT's search behavior.
