What Actually Moves AI to Recommend Your Hotel: The First Causal Evidence on Ratings, Reviews, and Replies
The short answer A hotel’s average guest rating is the single strongest reputation signal an AI travel planner responds to — and replying to reviews, the tactic most GEO vendors lead with, has no measurable effect on whether the machine picks you. That is the headline of the first pre-registered causal audit of LLM hotel selection, published in June 2026 by Baig, Gillani and Ali (arXiv, June 15, 2026). The researchers ran a randomized choice-based conjoint across twelve models — GPT-4o-mini, three Gemini versions, four Claude models, and four open-weight systems — presenting each assistant with sets of five hotel cards whose rating, review volume, review recency, management response, chain affiliation, price, eco-certification, and list position were independently randomized. Across more than 60,000 model calls, a top guest rating raised the probability of being the recommended hotel by 31.6 percentage points; a high price lowered it by 30.0 points; eco-certification added 11.6; review volume added 8.3; and a visible management response added 0.1 points — statistically indistinguishable from zero (arXiv, June 2026). ...