Study: LLMs give 100% monopoly to well-known brands—until a 0.1-star edge

A new research paper from Xi Chu and Yupeng Hou, posted on arXiv (2606.17443), investigates how LLMs handle brand competition when recommending products. Using skincare as a test case—where consumers rely heavily on brand reputation—the authors tested GPT-4o-mini, Claude Sonnet, and Gemini 3 Flash across three experiments.

Key findings reveal a "Conditional Monopoly": well-known brands get recommended 100% of the time (Innate Advantage Index = 10.0) when product specifications are identical. However, this dominance collapses if a competitor holds even a +0.1-star rating advantage. Authority-style marketing language, including fabricated clinical-evidence claims, breaks the monopoly at a Bias Surplus Value equal to +0.17 rating points—but each model responds differently. In a multi-brand GEO competition, a social dilemma emerges: when all brands adopt the same optimization strategy, individual payoff drops from +0.802 to +0.007, and non-participating brands receive zero recommendations. The paper suggests GEO should be studied both as a security risk and an emerging marketing practice.