Query Fan-Out Explorer: See What AI Search Really Asks

Enter a topic and we show the searches AI engines run behind the scenes to answer it: the queries Gemini executed through Google Search, the follow-ups Google AI Mode offers, what Perplexity searched, and what real people ask. Every row is labelled by evidence, and if you add your page you get a coverage score and a brief for the gaps.

Try: , , ,

A run takes 30 to 60 seconds. Results are estimates: no AI search engine publishes its real fan-out.

Evidence on Every Row

  • Observed

    A model actually ran this search: Gemini with Google Search or Perplexity, or Google AI Mode offered it as a follow-up.

  • Demand

    Real people search for it: People Also Ask, autocomplete, related searches or Reddit threads.

  • Cited

    Perplexity cites sources for this search, so you can see who gets the credit.

  • Model estimate

    Claude or OpenAI proposed it and nothing else confirms it yet.

Read the methodology: sources, label rules, stability weights and the price of every call.

Questions and Answers

What is query fan-out?

Query fan-out is how AI search engines answer a question: they split it into several related searches, run them in parallel and combine what they find. Google says AI Overviews and AI Mode may use it, and ChatGPT search, Gemini and Perplexity retrieve pages in a similar way.

Are these the real searches AI engines run?

Partly. Rows labelled Observed are searches a model actually ran: Gemini through Google Search and Perplexity through its own search, plus the follow-ups Google AI Mode offered. The rest are labelled Demand (real people search for them), Cited (an engine cites sources for them) or Model estimate. No engine publishes its full fan-out, so treat the result as a well-evidenced estimate.

What does the stability score mean?

It shows how many independent sources agree on a sub-query, from 0 to 100. A search Gemini actually ran that also appears in People Also Ask and in both models scores high; a sub-query only one model proposed scores low. The weights are on the methodology page.

How is the coverage score worked out?

If you add a page, we read its main content, find the passages closest in meaning to each sub-query and have a model judge whether each one is covered, thin or missing. The score is the share covered, with thin counting half, weighted by priority. It shows whether your page answers the sub-queries; it does not guarantee AI citations.

Why do I need to sign in?

Every run calls several paid AI and search APIs, so sign-in lets us give everyone five free runs a day and stop abuse. We only keep your email, your runs and their cost. Results are private unless you choose to share them.