Home / Query fan-out
Query fan-out: one prompt, many searches
The AI engine splits your question into parallel sub-queries, then merges them. Your keyword is no longer the query: it is one of the branches.
running shoes
First half-marathon, 90 kg, weak knees, €150 max: which shoes and where to buy?
- ↳ cushioned running shoes 90 kg
- ↳ best half-marathon shoes 2026
- ↳ running weak knees podiatrist advice
- ↳ running shoes under €150 in stock
crm
I run 12 reps, leaving Excel, €40/seat/month: which CRM?
- ↳ SMB CRM 12 users comparison
- ↳ CRM migration from Excel import
- ↳ CRM price per seat 2026
- ↳ GDPR CRM EU hosting
Definition
Google named the technique when it launched AI Mode (I/O 2025): the engine issues many parallel searches across subtopics. ChatGPT, Perplexity and Grok do the same under their own rules. It is a form of retrieval-augmented generation.
6–20sub-queries in AI Mode (Similarweb)
2,1 · 6,8fan-outs per prompt: ChatGPT · Grok (Peec)
≈ 31 %of ChatGPT prompts trigger web search (Superlines)
What it changes
- ChatGPT merges results with Reciprocal Rank Fusion: content present across several sub-queries rises (Peec).
- Sub-queries add words like “best”, “reviews”, the current year.
- They are non-deterministic: measure across prompt samples, not a single test.
- Definitions: little fan-out. Comparisons and lists: a lot (single-vendor data).
Map your fan-outs
- Write 10 realistic prompts per persona (see intent).
- Run each several times in AI Mode, ChatGPT, Perplexity.
- Log visible sub-queries and cited sources.
- Group them into facets: definition, price, comparison, alternatives, reviews, local.
- Create a section or page per facet and link them together.
Can sub-queries be seen?
Sometimes: AI Mode shows its search steps, and tracking tools collect them.
Keep exploring
Agentic promptsAgentic search promptsToolsKeyword and prompt toolsKeywordsThe keyword, 2026 editionAI enginesAI engines: what changes for SEO
Updated 27 October 2026