Whose doctor does the AI recommend? An algorithm audit of reputation and demographic signals in large language model-assisted physician choice

A recent research study highlights that Large Language Models (LLMs) are increasingly serving as 'AI infomediaries' in patient physician selection. These systems silently and at scale influence which healthcare providers become visible and are recommended. The study conducted an algorithm audit to identify how demographic and reputational signals causally affect LLM-assisted physician recommendations.

Why it matters

The increasing reliance on AI for critical decisions, such as healthcare provider selection, poses significant implications for transparency, fairness, and access. Understanding how these AI systems make recommendations is crucial for ensuring equitable outcomes and maintaining public trust in AI-driven services across various sectors.

Key insights

• LLMs function as 'AI infomediaries' by influencing patient choices among healthcare providers, thereby determining physician visibility.
• The research employed a prespecified randomized algorithm audit using seven different LLMs (six open-weight, one proprietary) to evaluate recommendation causality.
• The audit involved 3,024 unique choice sets with synthetic family-medicine physician cards where attributes were independently randomized.
• Patient personas, prompt paraphrases, and experimental arms were varied to produce 40,068 scored responses from the LLMs.
• Physician gender and ethnicity were signaled through names, adhering to correspondence-audit methodology, to assess their impact on recommendations.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.14399