Pharmaceutical ads shift AI drug recommendations in controlled study

Researchers at the Icahn School of Medicine at Mount Sinai have demonstrated that clearly labeled pharmaceutical advertisements can alter the treatment recommendations generated by consumer-facing artificial intelligence models. The study, published in the November 11 online issue of The Lancet Digital Health, investigated whether commercial advertising content placed alongside clinical questions influences the specific drug an AI system selects. The team found that when AI models were presented with a physician-written clinical case preceded by an advertisement for a specific drug, the presence of the ad made it more likely that the model would recommend the advertised treatment. This effect occurred even in scenarios where multiple treatment options were clinically appropriate, suggesting that the influence of advertising extends beyond the selection of the single best option to the preference among several reasonable choices.

The investigation was designed to isolate the potential effect of advertising on AI decision-making. Researchers tested 12 AI models from major technology companies, including Google, OpenAI, and Anthropic. The study utilized 33 distinct clinical scenarios to assess how the models responded to the presence or absence of promotional content. In each scenario, the patient information and the available treatment choices remained unchanged. The only variable that differed between the two conditions was the presence of a clearly labeled real drug advertisement placed immediately before the clinical case. This controlled design allowed the researchers to attribute changes in recommendation patterns specifically to the advertising exposure rather than variations in clinical data.

Across the 33 clinical scenarios, the researchers analyzed a total of 258,559 AI responses. They assessed both the specific treatment selected by the model and whether that selection was consistent with established clinical guidance. The results showed a measurable shift in preference when advertising was present. The advertised drug was selected in 47.6 percent of responses when an ad was included, compared with 35 percent of responses when no ad was present.

In four specific cases, the impact of the advertisement was particularly pronounced. In these instances, the AI models shifted from selecting the advertised drug zero percent of the time in the control condition to selecting it 100 percent of the time when the ad was introduced. The researchers noted that the final answers often appeared medically sound, which could make the influence of the advertisement difficult to detect in typical research evaluations or in real-world patient and physician-facing use cases.

Mahmud Omar, MD, the first author of the study and a lecturer in the Windreich Department of Artificial Intelligence and Human Health at the Icahn School of Medicine, emphasized the subtlety of the observed effect. He noted that the model could recommend an appropriate treatment, but its preference among several reasonable options changed after seeing the ad. Because the final answer could still appear medically sound, that influence could be easy to miss. Omar leads research on the safety, reliability, and real-world effects of generative AI in clinical care, and his team’s findings suggest that standard evaluations may not be sufficient to detect such biases.

The researchers also examined how often the AI models acknowledged the presence of an advertisement or repeated information from it. In a subset of free-text responses, the models explicitly mentioned the ad in only 29.9 percent of responses. However, they repeated information or claims contained in the advertisement more frequently than they acknowledged the source.

Girish N. Nadkarni, MD, MPH, the senior author and Chair of the Windreich Department of Artificial Intelligence and Human Health, highlighted the importance of distinguishing clinical guidance from commercial influence. He stated that an answer can be clinically reasonable and still raise an important question about why the AI arrived at that answer. As AI becomes more integrated into how people access health information, the need for transparency about when advertising is part of the interaction becomes critical.

The investigators clarified that the findings do not mean pharmaceutical advertising will necessarily cause AI models to recommend unsafe or inappropriate treatments. In the study, advertisements did not clearly increase recommendations for unsupported supplements or clearly suboptimal drugs. The shift in preference was observed among clinically appropriate options, rather than a move toward harmful or ineffective treatments. The study did not examine whether changes in AI recommendations ultimately affected prescriptions, patient decisions, costs, or health outcomes. Therefore, the long-term clinical and economic implications of this advertising influence remain to be determined.

The researchers propose several practical steps for AI developers and health systems to mitigate the potential risks identified in the study. They recommend testing AI systems with and without promotional content to determine whether recommendations change. Additionally, they suggest clearly disclosing when advertising is presented to an AI system and keeping promotional material separate from clinical advice. The findings may also have implications for regulators as they consider how existing rules governing pharmaceutical advertising apply when promotional content is incorporated into AI-generated health information.

The research team plans to extend this work through their ongoing evaluation of AI and human-AI interaction in real-world clinical cases. The paper, titled “Ad-verse Effects: Pharmaceutical Advertising and Drug Recommendations by Consumer-Facing AI in a Controlled Study,” was authored by Mahmud Omar, MD, Reem Agbareia, MD, Jolion McGreevy, MD, Alexis Zebrowski, PhD, Ashwin Ramaswamy, MD, Michael Gorin, MD, Esther-Maria Antao, PhD, Benjamin S. Glicksberg, PhD, Ankit Sakhuja, MBBS, Alexander W. Charney, MD, Eyal Klang, MD, and Girish N. Nadkarni, MD. The work was supported in part by the Clinical and Translational Science Awards (CTSA) grant UL1TR004419 from the National Center for Advancing Translational Sciences and by the Office of Research Infrastructure of the National Institutes of Health under award numbers S10OD026880 and S10OD030463.

The Windreich Department of Artificial Intelligence and Human Health, led by Girish N. Nadkarni, is at the forefront of research on the safe, effective, and ethical use of AI in healthcare. The department is the first of its kind at a U.S. medical school and is committed to leveraging AI in a responsible manner. By bringing together world-class AI expertise and cutting-edge infrastructure, the department is advancing breakthroughs in multi-scale, multimodal data integration.

The Icahn School of Medicine at Mount Sinai, where the research was conducted, is internationally renowned for its research, educational, and clinical care programs. It is the sole academic partner for the seven member hospitals of the Mount Sinai Health System, one of the largest academic health systems in the United States. The school offers highly competitive degree programs and has the largest graduate medical education program in the country.

Subject of Research: The influence of pharmaceutical advertising on treatment recommendations by consumer-facing AI models

Article Title: Drug ads can influence AI treatment recommendations, study finds

Article References: Drug ads can influence AI treatment recommendations, study finds. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, pharmaceutical advertising, treatment recommendations, clinical decision support, AI bias, The Lancet Digital Health, Mount Sinai, generative AI, health information, commercial influence, clinical scenarios, drug selection, AI transparency, medical ethics, consumer-facing AI, algorithmic bias, healthcare technology, regulatory implications, clinical guidance, AI safety