Fan-out queries: the searches AI engines run
optmz keeps the web searches each AI engine ran while answering, so you can see what it looked for, not just what it said.
What a fan-out query is
A fan-out query is a web search an AI engine writes for itself while answering a question. The engine rarely searches with the user's words. It rewrites the question into one or more shorter searches, often adding words like “2026”, “price”, “vs”, “official” or a `site:` filter for a brand's domain, and builds its answer from what those searches return.
That matters because ranking for the user's question predicts little. In 2026 studies, only about 8–12% of ChatGPT citations and 17–38% of Google AI Overview citations came from the top ten results for the original query. Being found for the fan-out searches is what counts.
What optmz shows
optmz shows the web searches each AI engine ran while answering, for every tracked question, wherever the engine exposes them. You can open a question and see the searches next to the answer and its cited sources, and see which modifiers keep coming back across runs.
- The searches per answer, kept with the answer text and its citations.
- Recurring searches across runs and engines for the same question.
- Fan-out gaps: searches the engines keep running whose answers leave the client out, which become fixes in the Action Center.
Which engines expose their searches
| Engine | Searches visible? |
|---|---|
| Gemini (API grounding) | Yes: the queries are returned with the answer |
| Claude (API web search) | Yes: each search query is returned |
| ChatGPT (API web search) | Usually: OpenAI says queries are returned “usually (but not always)” |
| Perplexity, Google AI Mode | Partly or not at all: the sources are visible, the searches mostly aren't |
When an engine doesn't return its searches, optmz doesn't guess them. More on how each engine searches is in the guides on ChatGPT and Google AI Overviews and AI Mode.