Commercial Drug Ads Significantly Bias AI Health Advice, Study Finds

Artificial intelligence models recommended advertised pharmaceutical treatments in 47.6% of responses when exposed to promotional material, compared to 35% without ads, according to research published in The Lancet Digital Health by the Icahn School of Medicine at Mount Sinai. The study highlights growing risks for consumers and tech platforms as everyday investors and households increasingly consult AI tools for medical information, demonstrating that clearly labeled promotional content can alter algorithmic recommendations.

Researchers tested 12 AI systems from major technology providers, including Google parent Alphabet Inc., OpenAI, and Anthropic, analyzing 258,559 total responses across 33 physician-written clinical cases. In four specific cases, introducing a drug advertisement caused the model's selection rate for the featured treatment to jump from 0% to 100%. Additionally, while models explicitly acknowledged the advertisement in only 29.9% of free-text responses, they frequently echoed specific claims and details from the marketing material.

The researchers noted that advertisements did not cause models to recommend unsafe or clearly suboptimal treatments, and the study did not measure direct effects on patient decisions, healthcare costs, or prescriptions. To safeguard clinical integrity, the study's authors recommended that tech developers and health systems test models against ad exposure, disclose promotional influences, and strictly separate advertising from medical advice. What to watch next: Watch for potential regulatory action as oversight agencies evaluate how traditional pharmaceutical advertising rules apply to AI-generated health advice.