Drug Ads Shift Consumer AI Treatment Recommendations, Study Finds
Researchers say the effect shows commercial content can nudge clinical advice, prompting calls for testing, disclosure, separation and regulatory review.
Overview
- The peer-reviewed study published Wednesday in The Lancet Digital Health found that clearly labeled pharmaceutical ads placed immediately before clinical queries increased the chance AI models recommended the advertised drug.
- Researchers tested 12 major consumer-facing models across 33 physician-written clinical scenarios and analyzed 258,559 AI responses to isolate the effect of a single variable: the presence of a real drug advertisement.
- Across responses the advertised drug was chosen in 47.6% of cases with an ad versus 35% without an ad, and in four scenarios the models shifted from never recommending the drug to recommending it in every response after seeing the ad.
- Models rarely acknowledged the ad when their answers aligned with it: in a subset of free-text replies only 29.9% explicitly mentioned the advertisement while they more often echoed its claims.
- Authors caution the study did not measure real-world effects on prescribing or patient outcomes and recommend developers and health systems test AI with and without ads, clearly disclose promotional content, separate ads from clinical advice, and expect regulatory review.