Online recommendation traditionally occurs after a user enters a platform, where the platform determines the candidate pool and the ranking shown to the user. LLM-based user agents enable a different recommendation process: a user specifies a need before choosing a platform, prompting platforms to compete for the user's attention, referred to as an agentic recommendation market.
In our controlled LLM-based experiments across three product domains, we find that this new recommendation setting creates tension between access and attention. Compared to traditional platform-centric recommendations, user-centric recommendations greatly expand the opportunity for relevant items to enter comparison; however, broader participation does not directly translate into effective exposure. Competition triggers strategic play from platforms: selectively positive explanations occupy 73-78% of top-ranked positions. When the user agent relates platforms' actions to subsequent user feedback, this share falls to 36-41%, while the likelihood of a user purchasing the relevant item increases. Thus, a user agent is more than a ranker over a larger pool of candidates: its querying, ranking, and feedback mechanisms determine who can compete, how scarce attention is allocated, and how earlier outcomes shape the evaluation of platforms, directly affecting user utility.
Designing agentic recommendations requires treating access, attention, and accountability as a joint mechanism design problem.
Blogger's Review: This research highlights how agentic recommendation markets reshape platform competition dynamics, emphasizing the importance of addressing user needs and platform strategies in the design of recommendation systems. Traditional mechanisms may fail to meet the diverse demands of modern users, necessitating more flexible and transparent designs in the future.