Key takeaways
- A query fan-out report lists the sub-queries an AI engine likely ran (or, in rare cases, actually ran) to answer a prompt. Most tools simulate this with an LLM guess; only a handful surface real backend queries.
- Don't treat every row as a keyword to target. Route each sub-query to one of three buckets: new page, section on an existing page, or discard. Building a thin page per row causes cannibalization, not visibility.
- Volume and phrasing change fast. Promptwatch's data shows ChatGPT's fanout queries dropped from an average of 117 characters to about 53 characters between December and April, and average queries per response swung from 2.15 down to 1.0, then back up after the site: operator rollout in August 2026.
- Engines behave differently: Perplexity mostly runs one query, ChatGPT runs a handful of short keyword-style ones, Google AI Mode fans out aggressively into 8-10+ sub-queries per prompt.
- Read the report as a diagnostic of what you're missing right now, not a permanent content architecture. Fan-outs regenerate differently between runs of the same prompt.
What you're actually looking at
A query fan-out report is a list of the narrower searches an AI system runs after it reads a single prompt. Ask ChatGPT, Perplexity, or Google's AI Mode a real question and none of them search your exact words. They break the question apart, run several searches in parallel against those pieces, and stitch the results into one answer. Google's own patent documentation (US11663201B2) calls this