Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems

Large language models that recommend products systematically favor well-known brands, but that advantage can be erased with authority-style marketing language, according to a study posted to arXiv on June 16, 2026 [1][2]. The research examined how brands compete inside LLM recommendation systems, using skincare products — a category where consumers cannot easily judge quality before purchase and must rely on brand reputation [1][2]. The authors tested three commercial models: GPT-4o-mini, Claude Sonnet, and Gemini 3 Flash [1][2]. When all products carried identical specifications, well-known brands were recommended 100% of the time, yielding an Incumbent Advantage Index of 10.0 [1][2]. That dominance dissolved when a competitor held less than a +0.1-star rating advantage [1][2]. Authority-style marketing language — including fabricated clinical-evidence claims — broke the monopoly at a Bias Surplus Value equal to +0.17 rating points, though each model responded differently [1][2]. The paper frames this dynamic as generative engine optimization, or GEO, an emerging practice that shapes market competition [1][2]. The study also identified a social dilemma: when all brands adopted the same optimization strategy, the individual payoff proxy fell from +0.802 to +0.007, and non-participating brands received zero recommendations in the tests [1][2]. The authors argue that GEO should be studied not only as a security risk but also as a marketing phenomenon [1][2]. The paper was submitted to arXiv, an open-access repository of electronic preprints that is moderated but not peer-reviewed [6]. As of November 2024, the repository was receiving about 24,000 articles per month [6]. The work appeared under the Computer Science > Artificial Intelligence category and was supported by arXivLabs, a framework that allows community collaborators to develop and share experimental tools directly on the arXiv site [4][5]. arXivLabs projects operate under guidelines that require partners to uphold values of openness, community, excellence, and user data privacy [4].