---
type: "Article"
title: "We pre-registered a query fan-out study. Answering the sub-queries did not predict AI Overview citation"
description: "A pre-registered test of 235 queries and 1,190 pages: once rank is held constant, query fan-out coverage adds nothing to predicting AI Overview citation."
resource: "https://www.laurelinlabs.com/blog/query-fan-out-study"
tags: ["query-fan-out-study", "ai-search-visibility"]
generated: { by: "laurelin-site-build/2026-09-02", at: "2026-09-02T12:00:00+00:00" }
status: "stable"
stale_after: "2027-03-02T00:00:00+00:00"
---
# We pre-registered a query fan-out study. Answering the sub-queries did not predict AI Overview citation

Published 2026-11-03. Author: [Piers Butler](/entities/piers-butler.md).

Not on this evidence. I captured the searches Gemini and Claude actually ran behind 235 real keywords with AI Overview citations on record and tested, within each query, whether pages answering those sub-queries were cited more often. Once organic rank, page length and domain rating were held constant, fan-out coverage added nothing to out-of-sample prediction (AUC change −0.003 against a pre-registered bar of +0.02), ranking for a sub-query added nothing, and guessed sub-queries did as well as observed ones. Rank was the strongest predictor. Sub-queries are good content research, not a citation lever.

## FAQ

### Does covering query fan-out sub-queries get a page cited in AI Overviews?

Not measurably in this test. Across 158 queries and 1,190 pages, adding fan-out coverage to rank, length and domain rating changed out-of-sample within-query AUC by −0.003 (interval −0.009 to +0.003), against a pre-registered pass bar of +0.02.

### How many searches does Gemini run behind a query?

Fewer than most tools suggest. With Google Search grounding, Gemini ran no search in 331 of 705 runs (47%), and the runs produced 2.35 distinct sub-query clusters per keyword, against 16.1 when a model is asked to guess.

### What did predict AI Overview citation in this study?

Organic rank for the head query was the strongest predictor (odds ratio 0.43 per standard deviation of log rank, p < 0.0001). Longer pages did a little better (odds ratio 1.30, p 0.015). Domain rating did not help (odds ratio 0.94, p 0.59).

Hub: [AI Search Visibility (AEO / GEO / LLMO)](/topics/ai-search-visibility.md)