Auditing Preferences for Brands and Cultures in LLMs

ArXi:2603.18300v1 Announce Type: cross Large language models (LLMs) based AI systems increasingly mediate what billions of people see, choose and buy. This creates an urgent need to quantify the systemic risks of LLM-driven market intermediation, including its implications for market fairness, competition, and the diversity of information exposure. Applied to Gemini, GPT, and DeepSeek across 10 topics spanning commerce and culture and than 2,000 questions, ChoiceEval reveals consistent preferences: U.