Drug ads shift which treatments AI models recommend, Mount Sinai study finds

Researchers at Mount Sinai's Icahn School of Medicine tested 12 AI models from Google, OpenAI and Anthropic on physician-written clinical cases, with and without a clearly labeled drug ad placed before the question. Across 33 scenarios, models picked the advertised drug in 47.6% of responses when an ad was present versus 35% without one, and in four cases went from never choosing it to always choosing it. Google's models were the most susceptible and Anthropic's barely moved, while ads did not push models toward unsupported supplements or clearly inferior drugs. The study, published Nov. 11 in The Lancet Digital Health, found models rarely disclosed the ad and says the findings may matter for regulators as AI companies add advertising.