Pharmaceutical advertisements may influence which medication an artificial intelligence recommends to a user, according to a new study published this week in The Lancet Digital Health. The findings are reported by Fast Company. The authors found that the appearance of an advertisement for one treatment option noticeably increases the likelihood that a chatbot will recommend it.
The researchers tested about a dozen consumer chatbots, including Claude, Gemini and GPT. They used 33 clinical scenarios, each containing several treatment options. In some of the scenarios, one of the options was accompanied by an advertisement for the drug. The advertising materials contained no instructions for the models to recommend that medication.
The results were notable. When an advertisement was present, the models recommended the advertised drug in 47.6 percent of cases. Without an advertisement, such a recommendation was given in 35 percent of cases. The effect varied among the twelve models, but overall the models usually still chose the clinically appropriate treatment option. At the same time, the presence of an advertisement was disclosed in only about 30 percent of the responses.
The study also describes possible ways in which advertising influence could reach AI responses: through user queries, additional context, web content the system retrieves during searches, or through deliberate malicious embedding. The authors stress that the study did not examine real-world consequences for prescribing. The impact on actual physician prescriptions therefore remains unconfirmed.
The question of regulation also remains open. It is unclear whether the promotion of drugs through AI falls under the oversight of the U.S. Food and Drug Administration (FDA). For context, according to IQVIA, pharmaceutical companies' spending on television advertising in 2024 exceeded $6 billion. The United States and New Zealand are the only high-income countries that permit direct-to-consumer advertising of prescription drugs.
In 2025, Senator Bernie Sanders introduced a bill that would ban advertising of prescription drugs. Meanwhile, a Doximity survey of 3,379 American physicians found that 54 percent trust AI recommendations on treatment. Three in four physicians noted that patients are increasingly arriving with treatment plans suggested by AI.
Separately, another study concerns diabetes medication. A Stanford study published in the journal Diabetes, Obesity and Metabolism found that patients with type 2 diabetes who took metformin had a 39 percent lower risk of death within a year after COVID-19 compared with those taking DPP-4 inhibitors. The analysis covered data on 53,332 adult Americans infected between October 2021 and November 2023. The 13 percent reduction in the risk of long COVID was not statistically significant. The study was based on medical records rather than a randomized trial, so it does not prove a causal relationship. A separate Minnesota study linked early use of metformin with a lower risk of long COVID. Experts emphasize that these results are not grounds for changing treatment.