Getting cited by AI is a method, not luck: the academic priority order is quotations, statistics, and cited sources

“GEO: Generative Engine Optimization” (arXiv 2311.09735, submitted November 2023, published at KDD 2024) built GEO-bench over 10,000 queries and measured visibility in AI answers as position-adjusted word count, testing nine rewrite methods.

Results relative to an unoptimised baseline: quotation addition about +41%, statistics addition about +31%, fluency optimisation about +28%, citing sources about +27%, technical terms about +18%; keyword stuffing about -8%.

A finding often missed: citing sources lifted visibility by roughly 115% for the fifth-ranked result but was negative for the first — evidence-building pays the most for challengers.

Brand teams whose content calendars do not yet treat “AI citation” as a metric.

Anyone deciding how to split budget between GEO and traditional SEO.

This is structurally good news for challenger brands: AI allocates attention not by domain authority but by whether a passage can be quoted directly. The writing itself is the optimisation — it costs budget only in editing time.

Direction is more stable than the percentages: the exact figures will shift with model iterations, but high factual density, clear structure, and sources keep working across model generations.