# Query fan-out

Query fan-out is the step where an answer engine turns one user question into its own set of background search queries and retrieves against those, rather than against the words the user typed. What the engine puts in those queries shapes the answer more than the pages it later fetches.

Fan-out sits between the question and the retrieval, which is why it is the hardest layer in this discipline to influence and the least measured. A publisher can control what a page says and whether a crawler can read it. Nobody outside the engine controls which queries the engine decides to run.

The published measurements are all on ChatGPT, and they are consistent about the size of the effect. An independent analysis of raw traffic from a ChatGPT Plus account found the model injecting brand names the user never typed into its own background queries in 21 of 27 tested conversations. Brands ChatGPT named in its own query reached the final answer 68.9% of the time, against 2.1% for brands that appeared only on retrieved pages — a roughly 33x gap.

The same analysis is the clearest published answer to a question practitioners ask constantly: how much retrieval converts to citation. Just 110 of 3,554 retrieved pages — 3.1% — earned a citation at all, and the rate fell from 5.2% at the first-listed domain to 0.3% from the sixth position on. Being fetched is not being cited.

Fan-out rules change without announcement, and the change is visible in the citation data before anyone explains it. Promptwatch attributed the August 2026 collapse in Reddit's ChatGPT citations to an August 8 shift in how the engine assembles these queries: use of the `site:` operator in fan-out queries rose from 0.37% to 16.8%, roughly forty-six-fold, which it reads as the engine asking named sites directly rather than searching the open web. Reddit's share of ChatGPT citations fell 86.4% in four days without Reddit changing anything.

Google describes a fan-out technique behind AI Mode, and the term is in common practitioner use for it. No comparable measurement of Google's query-writing step has been published, so the numbers above should not be carried across to AI Overviews or AI Mode — the mechanism is named on both sides, the evidence is not.

**What is query fan-out?** The step where an answer engine turns one user question into its own set of background search queries and retrieves against those instead of the words the user typed. Measurements published so far are all on ChatGPT, which injected brand names the user never typed into its own queries in 21 of 27 tested conversations.

**Why do AI engines search for things I didn't ask?** Because the engine reformulates the question before retrieving, adding terms and brand names it associates with the topic. That reformulation is where much of the outcome is set: brands ChatGPT named in its own query reached the final answer 68.9% of the time versus 2.1% for brands found only on retrieved pages.

**What percentage of retrieved pages actually get cited?** 3.1% in the one published measurement — 110 of 3,554 pages across 27 ChatGPT conversations — falling from 5.2% at the first-listed domain to 0.3% from the sixth position on. Retrieval and citation are different outcomes and the gap between them is large.

**Can you optimize for query fan-out?** Not directly, and nothing published shows a page-level lever that reaches it. The adjacent finding is that models search for brands they already know 3.2 times more often than unfamiliar ones, which points at brand familiarity rather than on-page work — suggestive, and not the same measurement.

Canonical: https://anythingengineoptimization.com/glossary/query-fan-out/
From the Anything Engine Optimization glossary — https://anythingengineoptimization.com/glossary/