AI Buying Agents Show Bias in E-Commerce

This research investigates the behavior of AI agents in online marketplaces, revealing several key biases and implications for agentic e-commerce. The finding that agents exhibit choice homogeneity, concentrating demand on a few 'modal' products, suggests a potential for market distortion. The instability of these preferences, with model updates causing drastic shifts in market share, highlights the volatility of agentic markets. The presence of strong position biases, even in text-only interfaces, indicates that agents are susceptible to ranking effects. The observation that agents penalize sponsored tags while rewarding platform endorsements raises concerns about fairness and transparency. The sensitivity of agents to price, ratings, and reviews, and the ability of sellers to manipulate agent behavior through description tweaks, further underscores the need for continuous auditing and regulation. These findings collectively suggest that agentic markets are fundamentally different from human-centric commerce, requiring new approaches to platform design, seller strategy, and regulatory oversight.

Transparency Disclosure: The analysis is based solely on the provided source content. No external information was used. The AI model used is Gemini 2.5 Flash.