Algorithm Audit Reveals Reputation Dominates AI Doctor Recommendations

A recent investigation published on arXiv (2608.14399) explores the role of large language models (LLMs) in guiding physician recommendations, characterizing these systems as 'AI infomediaries' that subtly influence patient decisions. The study conducted a prespecified randomized audit involving seven models—six open-weight and gpt-4o-mini—across 3,024 choice sets, three patient personas, nine paraphrased prompts, and nine experimental conditions, resulting in 40,068 evaluated responses. Utilizing synthetic family-medicine physician profiles with randomly assigned attributes, the research revealed that reputation signals are paramount: increasing a rating from 3.9 to 4.7 boosts choice likelihood by 31.4 percentage points, while escalating the fee from $90 to $190 decreases it by 20.0 percentage points. The study also examined demographic signals via names, but the abstract does not disclose those findings. These results underscore the significant impact of ratings and fees on algorithmic recommendations, raising issues of equity and accessibility in healthcare. The full study can be accessed at https://arxiv.org/abs/2608.14399.