Skip to content
Why GEO Why us Pricing
Get started free

Query fan-out generator

Query fan-out is the step where a generative engine turns one question into several retrieval queries, runs them all, and writes its answer from what comes back. This free tool models that decomposition for any question you type and labels every sub-query with the axis that produced it — so one question stops looking like one ranking contest and starts looking like the dozen-odd queries it really creates.

Query fan-out is also known as query expansion, query decomposition, sub-query generation, or simply the fan-out pattern. Google uses the term for AI Mode; the same behaviour shows up wherever an assistant searches before it speaks.

These sub-queries are modelled, not measured. They are what a published rule set says an engine would plausibly ask, given the words in your question. Nothing on this page observed an engine, and no line here is a query ChatGPT, Gemini, Perplexity or Claude was seen to issue. For fan-out that really happened, see measured fan-out below.

Ask it the way a buyer would. Runs in your browser: nothing is fetched, nothing is sent to a model, and no question is stored unless you ask for a share link.

Try: · · ·

A worked example

Take one ordinary buyer question: “best CRM for a small sales team”. It reads as a shortlist question, so the model fans it out along seven axes into 24 sub-queries. Every row below is the tool's own output, not a hand-picked illustration.

Modelled sub-queries for “best CRM for a small sales team”, by axis.
AxisModelled sub-queries
Who is on the shortlisttop crm 2026, crm comparison, crm recommendations
Head to headcrm comparison table, compare crm options, crm vs alternatives
What it costscrm pricing, how much does crm cost, crm pricing 2026, crm hidden costs
Fit for this situationcrm for a small sales team, is crm good for a small sales team, crm for a small sales team 2026, crm requirements for a small sales team
Objections and riskcrm problems, crm pros and cons, crm downsides, is crm worth it
Evidence people trustcrm reviews, crm reddit, crm case study, crm reviews 2026
Recencycrm 2026, best crm for a small sales team 2026

The useful part is not the list. It is that four of those seven axes — price, fit, objections and evidence — are questions most product pages never answer in their own words, which is why a page can rank respectably for the head term and still never be quoted.

The axes, and why each one fires

The model has to explain itself, because it cannot point at evidence. Every sub-query it emits is tagged with the axis that produced it, and every axis is a claim about how retrieval behaves that you can agree or disagree with:

Which axes fire depends on the shape of your question. A comparison splits its queries across both named things; a how-to question fans into steps, prerequisites and failure modes; a definition question never produces a pricing query, because nobody asks an engine what an idea costs.

Modelled fan-out and measured fan-out are not the same thing

This distinction matters more than anything else on the page, so it is worth being blunt about it.

This free toolMeasured fan-out in MentionBeat
Where the queries come fromA published rule set applied to your wordingThe retrieval queries the engine itself reported issuing
What it provesNothing. It is a hypothesis you can checkThat those queries were issued on that run
CoverageEvery question, instantlyOnly engines that publish their queries — the rest are reported as not reported, never as “no fan-out”
CostFree, unlimited, no accountA measured run

An engine that does not publish its retrieval queries has not told us it issued none — we simply did not see them. Collapsing those two into one number is the exact mistake this product exists to avoid, which is why the app states its fan-out coverage as “n of m engines reported” rather than quietly averaging the silent ones in as zero.

So: use the list below as a hypothesis about your category, then watch which sub-queries engines really issue for your own prompts. The paid product does not give you a better list. It tells you whether the pages you wrote against the list changed anything.

What to do with the list

  1. Tick what you already answer. Only count a sub-query as covered if a page of yours answers it directly, in its own words, above the fold of a section — not “it's in there somewhere”.
  2. Turn the gaps into H2s, not new pages. Most unticked lines become a heading with a self-contained two-sentence answer beneath it, on a page you already have.
  3. Check the page is readable at all. A perfect answer inside a client-rendered app is invisible — run a free page check, and confirm the crawlers can reach you.
  4. Mark the facts up. Prices, specs and Q&A extract far more reliably with structured data — build the JSON-LD.

For the wider picture of how Google's AI Mode assembles an answer out of background queries, read the Gemini and AI Mode playbook, or how AI Overviews choose their sources.

Questions people ask about query fan-out

What is query fan-out?

Query fan-out is the step where a generative engine turns one user question into several retrieval queries, runs them, and writes its answer from what comes back. It means a single question does not create a single ranking contest — it creates a retrieval surface of a dozen or more queries, and a page can be pulled into the answer by winning any one of them.

Are these the queries ChatGPT actually ran?

No. This tool is a model, not a capture. It applies a published set of rules to the words in your question and shows which rule produced each line. Nothing here observed an engine. Fan-out that was really issued can only come from the engines that publish their retrieval queries, which is what MentionBeat reports per run — including which engines did not report at all.

How many sub-queries does one question fan out into?

It varies by engine, by question and by run, and no fixed number is honest. Google has described AI Mode as issuing multiple background queries per question; the count is not published and is not stable. This tool caps its model at 24 so the output stays readable, which is a choice about the page, not a claim about any engine.

What do I do with the list?

Read it as a coverage checklist. Tick the sub-queries your existing pages already answer directly, and the unticked ones are the retrieval surface you are absent from. Most of them turn into an H2 with a self-contained answer under it, on a page you already have, rather than a new page.

Why this tool refuses to look more precise than it is. It would be easy to put a confidence percentage next to every sub-query, or to say “ChatGPT ran 14 searches for this”. Both would be invented, and inventing them would undercut the one thing this company sells, which is measurement you can check. So the tool shows its rules instead of its confidence, and the number of queries it emits is a display limit rather than a finding.