Here is a fun question for the AI era: if your chatbot doctor recommends a drug, is it because the drug is the best choice, or because someone paid for a banner ad? New research from the Icahn School of Medicine at Mount Sinai suggests the answer is a little bit of both.
Commercial pharmaceutical advertisements can significantly alter which medical treatments artificial intelligence models recommend, even when the systems clearly label the promotional content, according to the study. As more people turn to AI for health information, the Mount Sinai investigators examined how commercial advertising affects clinical questions. They published their findings in the November 11 online issue of The Lancet Digital Health.
How Drug Ads Sway AI Clinical Recommendations
The research team tested 12 AI models from leading technology firms, including Google, a subsidiary of Alphabet Inc. (GOOGL), OpenAI and Anthropic.
Investigators gave the models 33 physician-written clinical cases, either alone or immediately preceded by a clearly labeled real drug advertisement. Across 258,559 total responses, researchers kept patient details and treatment choices unchanged while varying only the ad's presence, evaluating both treatment selections and adherence to clinical guidance.
When an advertisement appeared, AI models chose the promoted medication in 47.6% of responses, compared with 35% without an ad. In four cases, models jumped from selecting the advertised drug 0% of the time to 100% once researchers introduced the ad.
And here is the sneaky part: in a subset of free-text responses, models explicitly acknowledged the advertisement in only 29.9% of answers, yet they repeated claims or details from the promotional material. So the ad was doing work even when the model was not saying so out loud.
Guardrails Needed for AI Health Advice
Before you panic, the investigators noted that drug ads did not clearly increase recommendations for unsupported supplements or clearly suboptimal drugs, meaning ads do not necessarily cause unsafe advice. The team also did not examine whether recommendation changes affected prescriptions, patient decisions, costs, or health outcomes.
To protect clinical integrity, researchers urged AI developers and health systems to test models with and without promotional content, disclose when systems receive ads, and separate advertisements from clinical advice. The findings also prompt regulators to evaluate how pharmaceutical advertising rules apply to AI-generated health information, as the team extends its ongoing evaluation of human-AI interactions in real-world clinical cases.
