AI-driven tools appear to be steadily gaining acceptance among physicians, according to a recent survey from The American Medical Association (AMA).

More than 4 in 5 doctors surveyed (81%) use AI in their practices, more than double the rate in 2023, results from the lobbying group’s 2026 Physician Survey on Augmented Intelligence found.

“I think much of that is likely ambient listening AI scribe tools…where, we’re having a tool that takes notes for the physician, so he or she doesn’t have to during the visit as well as generates after-visit summaries,” AMA CEO John Whyte, MD, MPH, told Medscape Medical News.

The survey — which examined perceptions about AI, its adoption and anticipated effects in medical practice across clinical, administrative, and patient-facing areas — also found that over three quarters of doctors surveyed now think AI enhances their ability to take care of patients, up from 65% in 2023. Respondents noted the biggest expected gains will be in work efficiency and diagnostic ability.

“AI has quickly become part of everyday medical practice,” Whyte said in a written statement. “Physicians see real promise in its ability to support clinical decisions and cut down on administrative burden.”

That said, “…For doctors to trust and use these tools, they must be safe, effective, and used responsibly, so they truly improve patient care,” Whyte added.

Prior to joining the AMA, Whyte was chief medical officer of WebMD from 2018 to 2025. (WebMD Health Corp is the parent company of Medscape).

Building on Prior Surveys

The survey, published in early March, builds on prior AMA surveys conducted in 2023 and 2024. This year’s survey was conducted over email. There were 2051 initial US physician participants while the final analytic survey consisted of responses from 1692 doctors who answered at least one nondemographic question.

Respondents represented a broad range of specialties, career stages, practice settings, and geographic regions. They had a median of 20 years in practice excluding residency and fellowship, 35 direct patient care hours/wk, and 26 physicians in their practice. Thirty-five percent of respondents were in the South, 21% in the West, 22% in the Midwest, and 22% in the Northeast. Specialties ran the gamut with the highest percentage (25%) in primary care.

Respondents were AMA and non-AMA members. However, the AMA did not explain how the survey population was selected.

Confidence in AI Tools on the Rise

In addition to an overwhelming majority of physicians (81%) reporting using AI in a professional context, the survey also found that the average number of use cases per physician doubled between 2023 and 2026 rising from 1.1 to 2.3 in that period. It also found that physicians most commonly use AI for medical research summaries and clinical care documentation.

A majority (70%) of physicians saw AI as a method to reduce work-related burnout by automating various tasks.

However, 40% of doctors said they were equally excited and concerned about AI with the top concerns being patient privacy and the integrity of the doctor-patient relationship. For broader adoption, most respondents highlighted the importance of data privacy (86%) and safety and efficacy validation (88%).

Physicians were supportive of patients using AI for medication and general health questions, but nearly half were against patients referring to AI tools to interpret pathology or radiology results.

One finding that stood out to Walter Sujansky, MD, PhD, president of health IT consulting firm Sujansky & Associates, was about how AI utilization has changed between 2023 and 2026. The largest increase in utilization was in the category of summaries of research and standards of care, rising from 13% in 2024 to 39% in 2026.

“My hypothesis is that this has a lot to do with the rapidly growing use of Open Evidence and similar tools that enable physicians to quickly inform themselves of the published best practices for diagnosing and treating specific patients,” Sujansky said in an email.

These tools may now serve an analogous role to previous research tools, such as UpToDate or Google, but are more efficient and effective, Sujansky said.

“There may now be substantial substitution of Open Evidence for those older research tools, accounting for the large uptick in the reported use of AI to summarize medical literature and practice guidelines.”

However, that may not be the kind of AI assistance that people think about when they think of AI use in healthcare, Sujansky added.

“They may be thinking about tools that exhibit some kind of human judgment or physician judgment actually making diagnoses or treatment decisions, ie, tools that directly assume some of the tasks of providing patient care. Those kinds of tools…are being used also but apparently not being adopted as broadly or quickly,” Sujansky said.

Results also showed that the second most common AI case use was creation of discharge instructions, care plans, and/or progress notes (used by 30% of physicians surveyed). Had that category been broken out further, Sujansky hypothesized that most of that utilization would have been concentrated in the creation of progress notes with ambient scribe programs that have grown rapidly in recent years.

‘Cautious Optimism’

The study was well done and provides more data on what’s actually happening in practice as opposed to anecdotes or marketing materials, said Spencer Dorn, MD, MPH, executive medical director of innovation at School of Medicine, The University of North Carolina at Chapel.

Overall, respondents were very positive about the direction AI tools in medicine were going but “at the same time they’re concerned about things like data privacy and the impact on the patient physician relationship,” Dorn said.

Whyte agreed and added that “there needs to be regulatory clarity regarding liability that influences adoption.” “Can we be sued if this tool gives you the wrong information? That’s why education is so important,” Whyte said.

The sentiment physicians have around AI is “trending positive,” Whyte said, “but there’s still this cautious optimism.”

“And that’s where the AMA is very interested in how do we help educate physicians how to utilize these tools? How do we help patients utilize these tools? And how do we ensure that they are validated from a safety, efficacy, and strong privacy protections standpoint,” Whyte said.

Limitations

Sujansky said that while the survey addressed a number of demographic and professional characteristics of respondents, the AMA did not address how representative the survey participants were of the actual physician population in the US. The AMA did not answer this question when Medscape asked.

Surveys are always limited in some ways in terms of who responds, Dorn said. People who respond to surveys about AI might be more interested in AI by default so there’s always a sampling question and a generalizability question.

Another issue is defining what utilization actually meant for the purposes of the survey, Dorn said.

“What does it mean that 81% use AI tools professionally? That’s probably a wide range of users.”

Dorn said he would have liked to see more “granularity” with the answers such as frequency of utilization. Does a doctor use an AI tool in a professional capacity once in a while or every day, for example?

Some of the categories of AI use that physicians were asked about also overlapped, Sujansky said — such as documentation of visit notes and creation of progress notes — and so it’s unclear exactly how much the use of ambient scribe programs alone may be contributing to the overall utilization of AI reported by the survey (which doubled from 2023 to 2026).

“Although it’s not something the survey was specifically designed to address, if the categories that could refer to the use of ambient-scribe tools are combined, the reported use of such tools could be over 50%,” Sujansky said in an email.

If that’s the case, the tools better be good and clinicians better be trained or otherwise know how to use them properly and review the output, Sujansky said.

“This is because those tools are known to not infrequently generate errors, usually small but sometimes significant, and they don’t always generate fully accurate progress notes and visit summaries,” he said.

A Useful Survey

The survey may not have been “methodologically rigorous,” but it’s useful, Dorn said. Doctors feel like AI-based tools are helping them provide better care.

“They see growing use over time. They generally have positive sentiments about it with a few concerns,” Dorn said.

Dorn advised several AI companies (eg, OpenEvidence, Doctronic, Roon). 

Sujansky reported having no disclosures regarding financial interests in healthcare AI companies.

Whyte reported having no disclosures.