← All research storiesHealth
Nearly a million U.S. clinicians asked an AI tool 271 million questions, mostly about how to treat patients
A JAMA study of one popular AI medical search tool shows how fast doctors, nurse practitioners and physician assistants adopted it, and what they ask it.
What the scientists did
OpenEvidence is an AI tool that clinicians use a bit like a search engine: they type a medical question and it writes an answer based on published medical research. Researchers at Harvard Medical School and Brigham and Women’s Hospital wanted to know who actually uses a tool like this, and for what.
They studied the tool’s records from January 2024 to June 2026, counting only accounts with a verified U.S. clinician ID number. There were far too many questions to read by hand, so they had a large language model sort a random sample of 38.8 million questions by purpose (for example, diagnosis or treatment) and by medical topic. Then they compared each group’s share of questions with its share of the workforce.
Words to know
Clinician
A health professional who directly cares for patients.
- Physicians (doctors)
- Nurse practitioners
- Physician assistants
Large language model (LLM)
A type of AI trained on huge amounts of text so it can read questions and write answers in everyday language.
- ChatGPT
- Claude
- Gemini
Cross-sectional study
A study that takes a snapshot of what people are doing, without testing whether it changes results.
- Counting what topics people search for
- Surveying how many students use a study app
What they found
About 952,000 U.S. clinicians, including about 556,000 physicians, sent roughly 271 million questions over 2.5 years. Monthly active users grew from 355 in January 2024 to 703,265 in June 2026, and the average active user went from 18 to 39.5 questions a month.
About half of all questions (51.9%) were about making clinical decisions. For physicians, treatment and management made up 33.1% of questions, while diagnosis made up 13.9%. Doctors within 10 years of graduating used it about 2.5 times more than their share of the workforce, while eye doctors, radiologists and orthopedic surgeons used it only about a quarter as much as their share. Many questions were outside the doctor’s own specialty, from 35% for cancer and blood specialists to 88% for emergency medicine.
Why it matters to you
Next time you see a doctor, there’s a real chance they have used an AI tool to double-check a treatment or look up the latest research. That can help busy clinicians keep up with new studies, but it also means the quality of these tools matters for everyone’s care. This study does not show whether AI made care better or worse, and it is not medical advice: questions about your own health belong with your doctor.
Where this could lead
Checking AI where it matters most
Hospitals and regulators could use usage data like this to decide which kinds of AI answers need the closest checking.
1–3 yearsStudies of patient outcomes
The researchers say the next step is to test whether using AI tools actually improves patient care.
1–3 yearsThese are possibilities the research points toward, not promises. Most early findings take years of testing before they reach everyday life, and some never do.
Why scientists care
AI is spreading quickly through medicine, but there has been little hard data on how clinicians really use it. Knowing that most questions are about treatment rather than diagnosis helps researchers decide what to test and where mistakes could matter most.
Keep in mind
This study counts questions; it does not show whether the AI’s answers were correct or whether patients did better. It covers one platform, the questions were sorted by an AI rather than by people, and three co-authors work for OpenEvidence (the company provided the data but did not fund the study).
How far along is the science?
- Peer-reviewedReviewed by independent experts for a journal before publication. Peer review is a quality check, not a grade of evidence strength: confidence still depends on study design, limitations, replication and the wider evidence.
- PreprintA manuscript shared publicly, often before or alongside journal review. Check its separate journal and review status (each story says what we know). Results may change.
- Conference talkPresented to other scientists at a meeting. Usually short and early; a full paper may come later.
- Agency reportPublished by a science agency such as NASA, NOAA or NIH. Reviewed internally, but not by an outside journal.
- Press releaseAn announcement from a university or company. Useful, but always check the research it describes.
How soon could it reach you?
- Could reach you soonCould show up in products, services or advice soon. Our rough editorial guess is about 1 to 3 years, not a forecast from the research.
- A few years outPromising, but needs more testing or engineering first. Our rough editorial guess is several years, with a lot of uncertainty.
- Long-term scienceBasic research that builds knowledge. Real-world uses, if any, are far off.
- Small stepA small update to earlier work. Useful to researchers, but not a change you would notice yet.
Every story links to its original source: the paper itself, or the conference program when no paper is out yet. Explanations are written with AI, usually from the paper’s abstract (each story says what it was written from), then checked against the source. We only use journals, recognized preprint servers, science agencies, scientific conferences and established science news outlets.