Research Investigates Advertising Influence on Drug Recommendations by Artificial Intelligence Models

A new study conducted by researchers at the Icahn School of Medicine at Mount Sinai has revealed that artificial intelligence models from major tech companies recommend advertised drugs at higher rates when matched with clinical queries.

The research, which tested 12 artificial intelligence models belonging to companies such as Google, OpenAI, and Anthropic, was published on November 11 in The Lancet Digital Health journal. The research team found that when a drug advertisement was added ahead of clinical cases written by a physician, this situation could alter the treatment recommended by the artificial intelligence and increase the likelihood of selecting the advertised drug. Across a total of 33 test results, the promoted drug was selected at a rate of 47.6 percent when the advertisement was present, compared to 35 percent when no advertisement was present. In four different cases, it was observed that the frequency with which models selected the advertised drug increased from zero to one hundred percent with the addition of the advertisement.

Dr. Mahmud Omar, the lead author of the study and from the Department of Artificial Intelligence and Human Health at the Icahn School of Medicine, stated that it was surprising how reasonable the responses appeared. Emphasizing that the model could suggest an appropriate treatment but changed its preference among several plausible options after seeing the advertisement, Omar noted that because the final response may appear medically sound, this effect might not be easily noticed in routine research and individual use scenarios.

The researchers stated that the findings do not mean drug advertisements lead artificial intelligence models to suggest unsafe or inappropriate treatments. It was noted that advertisements did not clearly increase the frequency of suggesting unsupported supplements or suboptimal medications. Furthermore, the study did not examine whether changes in artificial intelligence recommendations affected real-world elements such as prescriptions, patient decisions, costs, or health outcomes. Despite this, it was recorded that the findings could be important for regulators regarding the inclusion of promotional content in health information generated by artificial intelligence.