As travelers increasingly consult large language model (LLM) assistants for hotel bookings, these systems act as gatekeepers of hotel visibility, yet the factors driving their recommendations remain undocumented. We conducted a pre-specified algorithm audit using randomized choice-based conjoint analysis: across various personas, prompt templates, and twelve open-weight and proprietary models, assistants chose from five hotels with randomized guest ratings, review volume and recency, management responses, chain affiliations, prices, eco-certifications, and list positions.
We estimate the average marginal component effect of each signal on recommendation probability. Guest ratings and prices dominate (a top rating raises selection likelihood by 31.6 percentage points; a high price lowers it by 30.0), reflecting human valence-and-price primacy, while over-weighting eco-certification and neglecting management response. List position—a content-free artifact—causally shifts recommendations, valued at about $12 per night. Stated reasons imperfectly align with revealed weights. These findings ground generative engine optimization and the accountability of AI infomediaries in causal evidence.
Blogger's Review: This study provides a comprehensive audit of the decision-making logic behind AI hotel recommendations, highlighting the dominance of price and ratings. Notably, the overemphasis on eco-certification raises concerns. Understanding these mechanisms is crucial for enhancing user experience and ensuring transparency as AI technology becomes more prevalent.