Key facts
- A controlled study from the Icahn School of Medicine at Mount Sinai, published on 11 November 2026 in The Lancet Digital Health, tested whether a labelled drug advertisement changes what an AI model recommends.
- Researchers tested 12 AI models from major technology companies and analysed 258,559 responses to physician-written clinical cases.
- With an ad shown first, models were more likely to choose the advertised drug: about 48% of responses versus about 35% without an ad, as reported by Mount Sinai.
- The shift occurred mainly between options that were all clinically appropriate. Ads did not clearly increase recommendations of unsupported supplements or clearly suboptimal drugs.
What happened
The team gave each model the same clinical case with the same treatment options. Only one thing changed: a clearly labelled pharmaceutical advertisement was placed before the question in some runs and left out in others. The patient information and the treatment choices were identical.
Why it matters
More people now ask consumer AI tools about medicines, and some platforms are exploring advertising. The study suggests that an ad placed next to a clinical question can quietly tilt the answer, even when the answer is still medically reasonable. Because the tilt happens among acceptable options, it is hard for a user to notice.
Context: what the study did not show
The authors did not examine whether changed AI answers affected prescriptions, patient decisions, costs or health outcomes. The results describe model behaviour in a controlled experiment, not real-world prescribing. The senior author, Dr. Girish Nadkarni, said that clinical guidance and commercial influence need to be clearly separated when AI is used for health information.
The researchers suggest that AI developers and health systems test systems with and without promotional content, and clearly disclose when advertising is part of an interaction.
What it means for clinicians and students
- Clinicians: treat AI-generated drug suggestions as a starting point and check them against guidelines, especially when a tool is supported by advertising.
- Patients: ask where an AI tool’s information comes from and whether it carries advertising, and confirm medicine choices with a doctor or pharmacist.
- Pharmacology students: this is a concrete example of bias and conflict of interest in prescribing information, a theme that long predates AI.
Source: Mount Sinai press release (via News-Medical) on a peer-reviewed study in The Lancet Digital Health, 11 November 2026.
Source: Mount Sinai press release (via News-Medical), study in The Lancet Digital Health















