GEOHunter study traces a paid GEO network's placements into nine AI assistants
What this answers
- How much do paid GEO placements show up in AI assistants' citations?
- In one arXiv study of a single paid network on Chinese user-generated-content platforms, 4.5% to 20% of the retrievable pages that nine AI assistants cited were attributable to the network, depending on how strictly attribution was drawn.
- Do AI answers name clients of a paid GEO network?
- On eight of the nine assistants tested, the network's clients were named in 52% of answers, against 68% for the category leaders used as controls. The authors call these associations, not proof that the placements caused the mentions.
Researchers behind a new arXiv paper mapped one commercial generative engine optimization (GEO) operation across three Chinese user-generated-content platforms (Bilibili, Baijiahao and 360kuai), finding 1,985 analyzed accounts and 466,357 inferred promotional placements. The paper says its crawl reaches only accounts whose titles carry a searched client name, so these counts are lower bounds. On the network’s core Bilibili accounts, listicles rose from 16% to 57% of placements between 2024 and 2025.
The authors then asked nine AI assistants, from the US and China, buyer questions the network’s articles target, each paired with a category leader as a control. Depending on how strictly attribution was drawn, 4.5% to 20% of the retrievable pages the assistants cited came from the network. On eight of the nine assistants, the network’s clients were named in 52% of answers, against 68% for the category leaders.
The paper lists its own limits: a small question set (10 pilot and 14 census questions), one US vantage point, and engine-side results that are associations rather than effects of the placements.
Why it matters: Paid listicles on open publishing platforms are a measurable input to what AI assistants cite, and the authors say content-level checks cannot catch them because the signal sits at the account and platform level.
Glossary: GEO
Via arXiv GEO ↗
Posted to the wire November 20, 2026. Edited by Joe Balewski.