A drug ad is enough to move an AI's medical recommendation. That's the headline finding from a study published this week in The Lancet Digital Health: when researchers placed a pharmaceutical advertisement in front of a dozen consumer-facing large language models (AI systems that generate text, including chatbots like Claude, Gemini and GPT), the models recommended the advertised treatment 47.6% of the time, against 35% with no ad present.

That's a 12.6-point lift. And in most cases, the user never knew.

How the experiment worked

The research team ran 33 scenarios. Each one paired a clinical case description with a set of treatment options, and in some runs the researchers injected an ad for one of those options. The ad carried standard product information — what the drug does, what it's for — but it never explicitly told the model to pick that drug.

The models picked it anyway. The size of the shift varied across the 12 systems tested, which included versions of Anthropic's Claude, Google's Gemini and OpenAI's GPT, but the direction was consistent: ads nudged recommendations toward the advertised product.

Two details matter for executives watching this space.

• The ad-influenced answers were usually still clinically appropriate. The models generally picked one of the valid options, just not the same one they'd have chosen without the ad.
• The outputs disclosed the ad's presence only about 30% of the time. The influence was mostly invisible.

What does this mean for clinical guidance?

"An answer can be clinically reasonable and still raise an important question about why the AI arrived at that answer," study author Dr. Girish Nadkarni, professor of medicine and system chair of the Windreich Department of Artificial Intelligence and Human Health at Mount Sinai's Icahn School of Medicine, said in a statement. "As AI becomes part of how people get health information, we need to be able to distinguish clinical guidance from commercial influence and be transparent about when advertising is part of the interaction."

That's the crux. A recommendation can pass clinical muster while its reasoning was quietly shaped by commercial content. If nobody sees the ad, nobody can weigh it.

A $6 billion machine meets an unregulated channel

The commercial stakes are large. Pharmaceutical companies spent more than $6 billion on TV advertising alone in 2024, according to an IQVIA report, which stated that direct-to-consumer advertising "directly influences patient requests and prescription rates, driving real action in healthcare."

Now imagine that spending leaking into AI channels with no disclosure. The study flagged exactly this gap: models could encounter drug ads "through user prompts, application-added context, retrieved online material, or inadvertent or malicious insertion." And, unlike television or print advertising, which falls under U.S. Food and Drug Administration oversight, AI-mediated promotion "might not fit existing categories," the authors wrote.

A regulator designed for 30-second TV spots has no clean framework for a chatbot that has silently absorbed promotional content.

Where the regulatory pressure sits

Context matters here. Every high-income country except the U.S. and New Zealand bans direct-to-consumer prescription drug advertising. In the American market, prescription drugs consume more than 30% of commercial time during major networks' evening news programs, according to a 2025 release from the U.S. Senate Committee on Health, Education, Labor and Pensions.

Legislation is already moving. Last year, Ranking Member Senator Bernie Sanders introduced the End Prescription Drug Ads Now Act, co-signed by six other senators, which would ban such advertising outright, including on social media.

The Lancet study adds a new argument to that fight, though it carries one clear limitation: the researchers did not examine whether the measured effects translate into real-world changes in patients' requests or prescriptions.

Are doctors ready to catch this?

Probably not, and the data says so. A recent Doximity survey of 3,379 U.S. physicians found 54% expressed moderate or high trust in AI's ability to recommend treatment or management. Three-quarters of respondents said patients have become more likely to propose an AI-suggested diagnosis or treatment plan over the past six months.

Trust is rising. Defenses aren't. "Clinicians might defer to algorithmic recommendations even when they conflict with clinical judgement," the study's authors wrote, noting that an appropriate but ad-influenced suggestion "might provide little reason to question it."

The mechanism, then, is straightforward. Ads shift what AI recommends. The shifted answers still look sound. Doctors and patients increasingly trust those answers. And the ad behind the shift surfaces roughly one time in three.

For pharma, that's an influence channel with no FDA category. For everyone else, it's a transparency problem that current law hasn't reached.