Doctor-Driven AI

Nearly all physicians trust AI for medical literature search

Recent advancements in AI appear to have grown physician confidence in the technology.¹ The vast majority of physicians said they trust AI for a wide range of clinical tasks at least to a small extent, including medical literature search (94% of physicians), generating differential diagnoses (87%), and recommending treatment or management (86%). And more than three out of four physicians said they trust AI with identifying a potentially urgent or dangerous condition.

Physicians also remain optimistic about AI’s potential to further improve the way they work. When asked what excites them most about AI’s potential in their practice, physicians were most likely to mention improved work-life balance (65% of physicians) and a lighter administrative workload (59%). Many were also excited about its potential to promote better care by creating more time for patients (52%) and enhancing clinical decision support (51%).

*Physician respondents selected all that applied.

Physicians hold healthy skepticism of AI

Physicians have also maintained a healthy skepticism of the technology’s capacity.

More than half of physicians said they are concerned about AI’s accuracy and reliability, and three in four physicians have encountered a factual inaccuracy at least once in the past 12 months while using AI for work. In response to these limitations, two-thirds of physicians said that they independently verify AI-generated clinical information before using it all or almost all the time. Collectively, these experiences underscore the ongoing need for human oversight of AI in clinical settings.

Still, many physicians acknowledged that the overall quality of AI tools is going up. More than a quarter said that the effort they spend correcting or double-checking AI output has decreased compared with a year ago.

^*Physician respondents selected up to 5.

AI use has become a daily habit

Growing trust in AI appears to be translating into higher use among physicians.² Across the 3,379 U.S. physicians surveyed, 85% said they are using AI for work at least on a weekly basis. More than half (52%) said they use it every day.

The figures represent a significant upswing from 2025, when 51% of physicians reported using AI for work at least weekly and 37% at least daily. Indeed, when asked how their professional use of AI has changed compared with one year ago, 83% said it has increased somewhat or substantially.

^*Totals may differ slightly due to rounding.

AI use has increased across all age groups

In general, AI use trends higher among younger physicians. More than 90% of physicians 30 and under said they are using AI in their work at least weekly, compared with 89% in their 30s, 86% in their 40s, 80% in their 50s and 60s, and 66% in their 70s. AI adoption was also significantly higher this year than last year, increasing by at least 34 percentage points across all age groups.

Oncologists top all specialties in AI use

AI adoption remains high across medical specialties, exceeding 80% in most of the 20 specialties examined. Adoption of the technology is also distributed widely across primary care, surgery, and non-surgical specialists, reinforcing its versatility for diverse clinical settings.

Among the specialties studied, oncology had the highest AI usage rate. More than 93% of oncologists reported using AI at least weekly, including 65% who said they use it every day. Even in the two specialties with the relatively lowest rates, pathology and anesthesiology, about 70% of the physicians said they use AI at least weekly.

Two-thirds of physicians use AI for a task they used to do themselves

Two out of three physicians said they are now using AI to help them perform clinical or administrative tasks that they used to do on their own.

AI use for clinical tasks has jumped particularly high in the past 12 months. For example, 70% of physician AI users said they used AI for literature search this year, up from 51% the previous year. And 63% said they used AI for clinical decision support this year, up from 37% the previous year. There was also a nearly two-fold increase in using AI for interpreting medical images, lab results, or diagnostic tests.

Physicians continue to use AI to assist with writing, research, and administrative tasks as well. The most common applications this past year included professional correspondence (38% of physicians), patient education (34%), and insurance letters (32%).

^*Usage rate among physician AI users. ^† Physician respondents selected all that applied.

About half of physicians say AI has cut after-hours administrative work

Prior studies have indicated that many physicians spend a significant amount of time outside of regular work hours to complete administrative tasks like EHR documentation and paperwork, commonly referred to as “pajama time.”^3,4,5 In the current survey, practically all physicians reported spending at least some time on after-hours administrative work, including 26% who spend more than six hours a week.

AI is in position to help reduce this after-hours workload. In fact, in 2025, about a third of physician AI users (36%) said AI had already helped reduce their pajama time. By 2026, nearly half of physicians (46%) said it had.

The majority of physicians across all specialties surveyed—including primary care physicians, surgeons, and non-surgical specialists—estimate that AI could save them at least several hours in after-hours administrative work per week. About 41% of oncologists, 38% of general surgeons, and 23% of family physicians said AI could save them more than six hours per week. About one in 10 internists said AI could save them more than 20 hours per week.

AI is teaching physicians what they didn’t already know

More than simply using AI to support their work, many physicians are finding that AI is helping them advance their clinical knowledge and skills.

The vast majority of physicians (81%) said that using AI has taught them something clinically that they didn’t already know, such as an updated guideline, an alternative diagnosis, or a treatment option they hadn’t considered. And nearly one in four (22%) said that this occurs frequently. In addition, about two out of three physicians (65%) said that the availability of AI tools has helped with their skills development and knowledge. Although the benefits ranged widely for different specialties, from 82% of oncologists who said AI has helped to 48% of radiologists.

Physicians are concerned about AI-driven skill atrophy among trainees

While most physicians have found that AI has improved their clinical knowledge and skills, some remain concerned that younger physicians may overrely on it and, in turn, see their skills diminish.

Almost 61% of physicians said they are concerned about AI’s effect on the skills development of residents, fellows, or trainees they supervise. And 30% of physicians acknowledged that a top factor limiting their professional use of AI is concern over skill erosion. That concern is greatest among those 30 years old and younger (43% of physicians), which is the age group that includes the highest percentage of medical trainees.

^*Physician respondents selected up to 5 options.

Patient self-diagnosis is surging with AI

For decades, internet search engines have provided patients with access to medical information, allowing them to explore their symptoms online and self-diagnose.^6,7 AI seems to be driving that practice to new heights.⁸

About three in four physicians surveyed said that, over the past 6 months, patients have become more likely to propose a self-diagnosis or treatment plan from a generative AI tool. This includes 30% who said the frequency has increased dramatically.

Dr. AI is on track to outpace Dr. Google

AI chatbots also seem to have an edge over traditional online search engines at preparing patients for medical conversations. A third of physicians said that patients now arrive better informed with generative AI than they do with “Dr. Google.”

This advantage of AI applied across a wide variety of specialties. The five specialties most likely to report that their patients arrive better informed and ask better questions with the help of AI included primary care (internal medicine), surgery (urology), and several non-surgical major subspecialties (oncology, cardiology, gastroenterology).

AI misuse can pose real clinical risk to patients

At the same time, there remains plenty of room for improvement.⁹ About 27% of physicians said that patients are more confident in wrong conclusions about self-diagnoses due to generative AI. And 9% said that the most concerning AI-generated self-diagnosis they had encountered posed real risk if it had been acted on.

The results illustrate the potential for standard AI chatbots to either help patients or harm them, depending on the quality of the technology and the way patients use it.

*Due to rounding, numbers may not sum to 100%.

Over a third of physicians limit AI use due to legal and regulatory uncertainty

The rapid pace at which AI technology is progressing and being implemented into clinical practice has left a deficit of policy in its wake.^10,11 Almost 81% of physicians surveyed said that AI use is permitted or supported at their hospital or organization, but only 28% said their hospital or organization offered clear policies and approvals for its use.

That lag in AI rules and guidelines appears to be curbing physicians’ use of AI. When asked which concern most limits their professional use of AI, 34% of physicians cited legal and/or regulatory uncertainty.

Government hospitals are setting the pace on policy and approval

The presence of AI policies and approvals varied widely by type of practice.

More than half of physicians who worked at government hospitals and clinics (53%) said that their institution had established clear guidelines, compared with 14% of physicians working in private practice.

Physicians want to make the rules for AI in medicine

Physicians are among the primary users of AI in medicine and generally have a deep understanding of its effect on their work and patients.¹² So when asked who should make the rules for the appropriate use of AI, physicians overwhelmingly said that physicians should make the rules: 82% said it should be physicians, followed by 64% who said it should be medical societies, which are often led by physicians.

Less than half (45%) said that medical institutions should set the rules, and even fewer trusted federal regulatory agencies (32%), state governments (20%), or AI developers and companies (10%) with that responsibility.

*Physician respondents selected all that applied.

Methodology