What is query fan-out?
When you ask an AI assistant a question, it rarely answers from memory alone. It fans out: it runs several web searches behind the scenes, reads what ranks, and cites those pages in its answer. Those intermediate searches are the fan-out queries — and they decide who gets cited. Rank for the fan-out query and you're in the answer; miss it and you're invisible, however good your content is.
Where the data comes from
Four engines expose their fan-out: ChatGPT, Gemini, Perplexity and Claude. Locuvox captures the queries automatically as your scheduled checks run — nothing to configure, and no extra checks consumed. Engines that don't expose their searches (Google AI Overviews, AI Mode, Grok) don't appear here.
Reading the page
- Search query — the literal search the engine ran. Badges mark queries containing your own brand, a tracked competitor's name (set competitors via the "competitor" domain tag on Sources), and queries first seen in the last week.
- Engines — which engines ran it.
- Seen — how many times it's been captured, with a trend arrow once there's two weeks of history.
- You cited — of the answers this query helped generate, how many cited your domain. "Not cited" is the important flag: the query is driving answers in your category and your content isn't in them.
Click any query for the detail view: sightings per day, per-engine counts, and the prompts it appeared for. The whole list exports to CSV.
Read vs cited (ChatGPT & Perplexity)
For ChatGPT and Perplexity — the engines that expose their reading list — clicking a query also shows every page the engine read while answering, split into two lists:
- Cited in the answers — pages that made it into the response. This is the bar your content has to clear.
- Read but not cited — pages the engine consulted and passed over. If your page is here, you're ranking well enough to be read but the content isn't winning the citation — a content problem, not a ranking problem. If your page is in neither list, it was never read: rank for the fan-out query first.
The Domains tab
The read set rolled up per domain: how often each domain gets fetched by the engines' searches versus how often it's actually cited. The "fetched, not cited" column is the one to watch — a domain the engines keep reading but won't quote means the content isn't giving them what they cite: plain-text facts, numbers, quotable lines. If that domain is yours, that's your editing brief.
The Shopping tab (ChatGPT)
On buying-intent prompts, ChatGPT has two product tiers: a product card (from the merchant feed, linking straight to the seller) and an inline mention (the editorial text tier). Locuvox captures both per merchant, so a retailer can see whether they're winning the conversion-grade card or only being mentioned in passing — and which competitors hold the cards.
What to do with it
- Sort by Seen and scan the top queries marked not cited — that's your highest-leverage content list.
- Check the competitor-badged queries: AI is actively researching rivals on your prompts.
- Hand the CSV to whoever writes your content — each query is a page or section brief with intent built in. Recommendation briefs on the Insights page already weave these queries in.
Why numbers start small
Capture began the moment the feature launched, so a new project shows only what its recent checks collected. The picture compounds with every scheduled run — trends and "new query" badges switch on automatically as history builds.