Final up to date: November 6, 2026. Based mostly on a JAMA research revealed October 15, 2026.
Practically 1,000,000 U.S. clinicians, greater than half of them physicians, have left a document of what they ask AI, and a brand new JAMA research simply analyzed it.
The brief reply: U.S. clinicians, together with 556,200 physicians, requested OpenEvidence, an AI medical data platform, 271 million questions between January 2024 and June 2026. Amongst physicians, therapy and administration questions (33.1%) outnumbered prognosis questions (13.9%) by greater than two to at least one. Nurse practitioners and doctor assistants requested probably the most questions per person, and early-career physicians used the platform properly above their share of the workforce, based on a research revealed in JAMA on October 15, 2026.
The research fills a niche. Its authors word that empirical proof on how medical doctors and different clinicians use AI instruments has been restricted. It coated everybody on the platform, and the remainder of the customers included nurse practitioners and doctor assistants, so some figures beneath describe all clinicians and others describe physicians solely. Every is labeled.
Researchers at Harvard Medical College and Mass Normal Brigham performed the research with co-authors from OpenEvidence and different establishments. Three co-authors had been OpenEvidence workers, and the corporate provided the information, which readers can weigh when studying the outcomes.
Disclaimer: Whereas these are basic solutions, it’s vital to conduct thorough analysis and due diligence when deciding on AI instruments. We don’t endorse or promote any particular AI instruments talked about right here. This text is for academic and informational functions solely. It’s not supposed to offer authorized, monetary, or medical recommendation. All the time adjust to HIPAA and institutional insurance policies. For any selections that affect affected person care or funds, seek the advice of a certified skilled.
Key numbers at a look
| Measure | Discovering |
|---|---|
| Examine interval | January 1, 2024 to June 30, 2026 |
| Clinicians (all) | 952,008, together with 556,200 physicians |
| Questions submitted (all clinicians) | 271 million |
| Month-to-month lively customers (all clinicians) | 355 (January 2024) to 703,265 (June 2026) |
| Questions per lively person per thirty days (all clinicians) | 18.0 to 39.5 |
| Questions per lively person per thirty days (physicians) | 37.1 |
| Prime doctor function | Remedy and administration (33.1%); prognosis (13.9%) |
| Questions exterior personal specialty (physicians) | 35% (oncology and hematology) to 88% (emergency medication) |
Supply: Patel et al., JAMA, October 15, 2026.
What did the JAMA research take a look at?
The researchers analyzed accounts verified by way of Nationwide Supplier Identifier numbers, so each person was a licensed clinician. In all, 952,008 clinicians, together with 556,200 physicians, submitted 271 million questions over 30 months.
No one can learn 271 million questions by hand. The staff took a random pattern of 38.8 million, constructed to signify totally different person teams, and used a big language mannequin to label every query by medical function and matter. The class percentages come from an AI classifier, not from individuals reviewing every query.
How briskly is clinician AI use rising?
In January 2024, 355 clinicians used the platform in a month. By June 2026, the month-to-month rely was 703,265. Common questions per lively person rose from 18.0 to 39.5 every month, so extra clinicians joined and people utilizing it requested extra. Becker’s Hospital Evaluation reported that month-to-month retention climbed from about 70% to 90%.
Who makes use of medical AI probably the most?
Nurse practitioners and doctor assistants every averaged 44.5 questions per lively person per thirty days. Physicians averaged 37.1.
Profession stage confirmed a bigger hole. The research reviews a “use ratio,” which compares a gaggle’s share of questions with its share of the nationwide workforce. A ratio of 1.0 means use in proportion to measurement. Physicians inside a decade of commencement scored 2.54, and residents and fellows scored 1.84. Becker’s reviews the identical sample in uncooked shares: physicians inside a decade of commencement had been 24.3% of the doctor workforce and despatched 61.8% of doctor questions.
Specialty mattered too. Oncology and hematology (2.00), inner medication (1.88), and household medication (1.80) ran above their workforce share. Ophthalmology (0.23), radiology (0.24), and orthopedics (0.24) ran far beneath, at roughly 1 / 4 of their share. The research describes these patterns however doesn’t check why they exist.
What do clinicians ask AI about?
Most questions (51.9%) involved medical decision-making. Amongst doctor questions, therapy and administration made up 33.1% and prognosis 13.9%, so physicians requested about therapy and administration greater than twice as usually. The authors conclude that use was “often for therapy selections somewhat than diagnostic duties.”
Becker’s reviews that the authors see a niche right here: most medical AI evaluations check diagnostic accuracy and data recall, whereas therapy choice, treatment steerage, and proof synthesis stay largely unmeasured. The paper doesn’t clarify why clinicians ask about therapy extra.
How usually do physicians ask exterior their very own specialty?
It relies upon closely on the sector. Amongst oncologists and hematologists, 35% of questions fell exterior their specialty. Amongst emergency division physicians, 88% did. The authors warning that a part of this sample might replicate how broad the platform’s content material is, because it has content material partnerships with a number of main medical publishers. The research doesn’t clarify the remaining.
What can’t this research inform physicians?
A number of limits matter earlier than anybody repeats these numbers.
The research measured customers, questions, share relative to the workforce, and the aim and matter of questions. It didn’t measure accuracy, security, or affected person outcomes. A software could be extensively used and nonetheless be proper or improper in any given reply, and this research can’t separate the 2.
It covers one platform, and the authors word that patterns might not carry over to instruments with totally different content material entry. The information additionally finish in June 2026.
The authors’ ties matter as properly. In line with the JAMA disclosures, a co-lead creator was a salaried OpenEvidence worker, and OpenEvidence supplied the information and contributed to the evaluation. Becker’s reviews that two different co-authors, together with the corporate’s chief medical officer, had been additionally workers. The paper went by way of JAMA’s peer assessment, and the disclosures are public, so readers can weigh them alongside the findings.
Becker’s additionally notes that the authors deal with their counts as decrease bounds, as a result of the research included solely questions typed into the platform’s most important interface.
How ought to physicians learn a utilization research?
Research about AI in medication reply totally different questions. A utilization research asks who makes use of a software, how usually, and for what. An accuracy research asks how usually the software is correct. An outcomes research asks whether or not sufferers did higher.
This paper is a utilization research, and a cross-sectional one, which describes patterns in knowledge collected over a interval. It could present that early-career physicians use the platform extra. It can’t present that utilizing it modified their care, studying, or careers. JAMA ran an editor’s word with the paper, noting that these instruments have an effect on how clinicians study and weigh proof, not solely how they appear issues up.
Three questions assist with any research of this kind. What precisely was measured? Who provided the information and helped analyze it? Does it say something about high quality, or solely about recognition?
Unlock the Full Energy of ChatGPT With This Copy-and-Paste Immediate Method!
Obtain the Full ChatGPT Cheat Sheet! Your go-to information to writing higher, sooner prompts in seconds. Whether or not you’re crafting emails, social posts, or shows, simply observe the method to get outcomes immediately.
Save time. Get readability. Create smarter.
The underside line
AI medical reference instruments are actually a part of on a regular basis work for a lot of U.S. clinicians, physicians included, and use remains to be rising. Remedy questions outnumber diagnostic ones, early-career physicians and nurse practitioners and doctor assistants use the platform most, and lots of questions attain past a clinician’s personal specialty.
The research leaves accuracy and affected person outcomes open, and people want other forms of analysis. Till then, a utilization research like this one is greatest learn as a measure of adoption, not of high quality.
What do you assume? Do you agree with what the analysis discovered? Tell us within the feedback!
Obtain The Doctor’s Starter Information to AI – a free, easy-to-digest useful resource that walks you thru sensible methods to combine instruments like ChatGPT into your skilled and private life. Whether or not you’re AI-curious or already experimenting, this information will prevent time, stress, and perhaps even somewhat sanity.
Need extra tricks to sharpen your AI abilities? Subscribe to our publication for unique insights and sensible recommendation. You’ll additionally get entry to our free AI useful resource web page, full of AI instruments and tutorials that will help you have extra in life exterior of medication. Let’s make life simpler, one immediate at a time. Make it occur!