What Is Query Fan-Out (and Why AI Cites Pages That Do Not Rank)

Query fan-out is how AI search engines rewrite one question into several hidden sub-queries, retrieve each separately, then merge the results into a single answer. It is why a page that ranks nowhere for the head keyword can still get cited, and why a page ranked number one can still be invisible. What matters is whether you own a passage that answers one of the hidden sub-queries, not the keyword the buyer actually typed.

What fan-out looks like from the inside

A buyer asks "best CNAPP for a mid-market fintech." The engine does not search that sentence verbatim. It rewrites it into candidate sub-queries: which vendors offer CNAPP, which of those support fintech compliance needs, which have strong Kubernetes runtime coverage, and what pricing looks like at that company size. Each sub-query runs its own retrieval pass. ChatGPT in particular runs a second round per candidate: round one builds the shortlist, round two fetches each vendor individually for pricing, coverage, and reviews. You need to win both rounds. Round one gets you named. Round two decides how you get described.

Google AI Overviews and AI Mode work the same way against Google Search, which is why keyword-rank thinking breaks down so badly on these two surfaces specifically.

This is why keyword-rank thinking breaks

You can rank number one and be invisible. You can rank nowhere and be cited. The old model, that a top-ranking page earns the citation, does not hold once retrieval happens per sub-query instead of per head term.

Ahrefs' analysis of AI Overview citations across roughly 863,000 keywords found only about 38% of cited URLs came from the page's own top 10 organic results, and roughly 37% of cited URLs did not appear in the top 100 organic results for that query at all. In Google's AI Mode specifically, 31% of cited URLs came from organic positions 11 through 100, and another 31% came from beyond position 100. Your organic rank on the head term and your odds of citation are, for a large share of queries, two separate games.

The practical consequence: a page that thoroughly answers one narrow sub-question, "Kubernetes runtime coverage for a fintech-sized CNAPP deployment," has a real shot at citation even with zero rank on "best CNAPP," because that sub-query is a distinct retrieval event the head-term ranking never touches.

How to reconstruct your own fan-out footprint

Different engines expose this differently, and none of them hand you a clean log.

Google AI Mode. Ask it directly: "What sub-questions did you research to answer that?" It is a reconstruction of what it did, not a verbatim log, so treat it as directional rather than exact.

Google AI Overviews. Expand the citation panel, then run the same head query in normal Google search and compare the two citation lists. The gap between them is your fan-out footprint: URLs that show up in the AI Overview but nowhere in the top organic results got there through a sub-query the head-term ranking never surfaces.

ChatGPT and Perplexity. Ask directly for every source used, then push further: "which specific source did each claim come from." Perplexity lists sources by default with per-claim attribution, which makes this the easiest engine to reverse-engineer.

What this means for content strategy

Optimizing only the head prompt while ignoring the fan-out is the single most common wasted effort in a GEO program. Two practical shifts follow.

Write for the sub-question, not just the page title. Every H2 on a page should be a self-contained answer to one narrow question a fan-out pass might generate. A reader, or a retrieval system, arriving at that H2 alone should get a complete answer without needing the rest of the page.

Specificity beats breadth. "Best CNAPP for a Kubernetes-heavy fintech under 500 employees" is worth more coverage effort than "best CNAPP." It has less competition, it matches a real fan-out sub-query almost verbatim, and the mechanism actively rewards content that answers narrow questions over content that gestures at a broad one. Building a prompt universe is how you find which narrow sub-questions your buyers actually generate, instead of guessing.

Fan-out and the freshness myth

Fan-out retrieval does not automatically favor new content. In a large sample of AI Overview citations, the median cited page skewed older than most teams assume, which means a well-structured page from a year ago can still win a sub-query it answers cleanly. Freshness helps, but structure and specificity carry more weight than a recent publish date alone.