Study Finds A.I. Chatbots Frequently Provide Incorrect Health Advice

Experiment shows chatbots no better than Google at guiding users to correct diagnoses or next steps.

Published on Feb. 9, 2026

Got story updates? Submit your updates here. ›

A new study published in Nature Medicine found that A.I. chatbots, which have become a major source of health information, frequently provide incorrect medical advice. The experiment showed chatbots were no better than general online searches at helping participants identify the right course of action or diagnose medical conditions based on detailed scenarios. Researchers found the chatbots struggled with distinguishing urgent from non-urgent situations and sometimes returned fabricated information.

Why it matters

As A.I. chatbots have become increasingly popular for answering health questions, this study raises concerns about their ability to provide reliable and safe medical guidance. The findings suggest these tools may pose unique risks compared to traditional online health research, potentially leading users toward inappropriate or even dangerous next steps.

The details

The study, led by researchers at the Oxford Internet Institute, involved over 1,200 British participants with no medical training. They were given detailed medical scenarios and asked to chat with commercial chatbots like OpenAI's ChatGPT to determine the appropriate next actions. The participants chose the 'right' course of action less than 50% of the time and identified the correct conditions only about 34% of the time - no better than a control group using general online research. The researchers found issues like chatbots struggling to distinguish urgent from non-urgent situations and sometimes providing fabricated information, as well as the models being highly sensitive to small changes in how questions were phrased.

  • The study was published on February 9, 2026.

The players

Adam Mahdi

A professor at the Oxford Internet Institute and senior author of the new Nature Medicine study.

Andrew Bean

A graduate student at Oxford and lead author of the paper.

Ethan Goh

Leads the A.I. Research and Science Evaluation Network at Stanford University.

Robert Wachter

Chair of the department of medicine at the University of California, San Francisco, who studies A.I. in health care.

Danielle Bitterman

Researcher who studies patient-A.I. interactions at Mass General Brigham.

Got photos? Submit your photos here. ›

What they’re saying

“Medicine is not like that. Medicine is messy, is incomplete, it's stochastic.”

— Adam Mahdi, Professor, Oxford Internet Institute (Nature Medicine)

“Is it really the user's responsibility to know which symptoms to highlight, or is it partly the model's responsibility to know what to ask?”

— Andrew Bean, Graduate Student, Oxford University (Nature Medicine)

“Very, very small words make very big differences.”

— Andrew Bean, Graduate Student, Oxford University (Nature Medicine)

What’s next

The researchers noted that A.I. companies are working to improve chatbots' ability to ask follow-up questions and better distinguish urgent from non-urgent situations, but more research is needed to evaluate the latest models.

The takeaway

This study highlights the significant limitations of current A.I. chatbots in providing reliable and safe medical advice, raising concerns about their use as a primary source of health information. It suggests users should be cautious about relying on chatbots for anything beyond general wellness tips and should always consult qualified medical professionals for serious health issues.