Generalization of AI-driven commercial persuasion beyond eBooks

Determine whether the large increases in sponsored-product selection produced by conversational large language model shopping agents in the eBook domain generalize to higher-stakes purchases (such as electronics or financial products) and to more commoditized goods.

Background

The paper reports two preregistered experiments (N = 2,012) comparing traditional search placement with conversational LLM agents instructed to promote sponsored products during an eBook shopping task. Persuasion by LLMs substantially increased selection of sponsored items (e.g., 61.2% in the Chat–Persuasion condition versus 22.4% under Search–Placement), while most users failed to detect promotional steering.

The authors identify limitations, including that the study examines only a single product domain (eBooks). They explicitly flag uncertainty about whether these effects would extend to higher-stakes categories (e.g., electronics or financial products) or more commoditized goods.