GEO content scores barely predict AI citations, study finds
What this answers
- Do GEO content-scoring tools actually predict which content AI engines cite?
- Not well, per a new study — a deterministic content score's within-query correlation with actual citations was just 0.11, positioning it as a content-quality filter rather than a citation predictor.
- Do the 'GEO lever' techniques identified in 2023 studies still boost AI citations?
- No — re-testing the original 2023 effect sizes on ten modern AI engine families found none of the levers moved citations, per the study.
A new arXiv preprint validates a deterministic content score built as a cheap proxy for AI-citation likelihood, using adversarial falsification tests — negative control, dose-response, bounded amplification — rather than assuming the proxy holds. Re-testing the field’s original 2023 citation-lever effect sizes on ten modern AI engine families found none of the levers moved citations, and recalibrating to current data strips the score of its lever-responsive components entirely. On a 500-source adversarial benchmark, amplifying the surviving levers gained an attacker at most 6 points, and the score’s own citation-predictive power was weak — a within-query Spearman correlation of 0.11 — positioning it as a content-quality filter rather than a citation predictor, per the authors.
Why it matters: It's evidence that the citation levers the GEO industry's early tooling relied on have expired on today's engines, and that content scores marketed as citation predictors are, at best, quality filters.
Glossary: GEO
Via arXiv GEO ↗
Posted to the wire September 8, 2026. Edited by Joe Balewski.