The AI Exam Room: What 1,692 Physicians Actually Want (and Fear) in 2026

Maya: Over 80% of physician respondents currently use AI in a professional context, which is exactly double the share reported in 2023.

Sam: That is a staggering jump in just three years. Welcome to AI in Medicine - Smart Summaries. Just a quick note right at the top: our voices are AI-generated, and today we are bringing you a summary of a completely public document.

Maya: Today we are unpacking the 2026 Physician Survey on Augmented Intelligence. It was published in March 2026 by the American Medical Association, specifically their Center for Digital Health and AI.

Sam: And the reason we chose this document is the scale. They surveyed a total of 1,692 physicians. This is not a tiny sample of early adopters in Silicon Valley. This is a massive cross-section of the actual medical workforce.

Maya: Exactly. The survey was fielded from January 15th to February 2nd, 2026. And the respondent profile proves your point. We are looking at a median of 20 years in practice, and a median of 35 direct patient care hours per week.

Sam: These are busy, experienced clinicians. 35% are in the South, 22% in the Midwest, 22% in the Northeast, and 21% in the West. It covers solo practices, hospital settings, and group practices. So if you are sitting in a boardroom trying to figure out what doctors actually want from AI, this is your roadmap.

Maya: Let us start with the sheer volume of adoption. As I mentioned, over 80% use AI professionally. But what is really interesting is how their use is expanding. The average number of use cases per physician increased from 1.1 in 2023 to 2.3 in 2026.

Sam: So they are not just dabbling with one single tool anymore. They are layering multiple applications into their daily routines. And the uncertainty is vanishing. The report notes that uncertainty about available AI tools fell from 18% in 2024 to just 9% in 2026.

Maya: Meanwhile, the group claiming no reported AI use fell from 34% down to 19% over that exact same period. The ambiguity is essentially gone from the clinic.

Sam: I was also struck by how deeply embedded this is in their personal lives. Over three-quarters of physicians use an AI system at least once a week for personal activities, and a third are using it daily.

Maya: The document explicitly names things like AI-assisted internet search, customer service chatbots, image generators and AI chatbots like ChatGPT in that personal use category. That personal familiarity is bleeding directly into their professional expectations.

Sam: So what exactly are these 1,692 doctors doing with AI at work? What is the number one use case?

Maya: The top application right now is summaries of medical research and standards of care. Nearly 40% of respondents are using it within their workflows. That is 26 percentage points higher than 2024.

Sam: Wait, really? 26 percentage points in two years? That is a massive spike for a single use case.

Maya: It is. The report says it has quickly become the most common AI use case incorporated into practice. It makes sense from a risk perspective. Summarizing existing standards of care carries a different weight than generating a fresh diagnosis for a patient in the room.

Sam: But the clinical applications are coming fast. The survey asked physicians who considered certain use cases relevant when they expect to incorporate them. Let us look at the near-term pipeline.

Maya: The numbers here are huge for medtech companies to pay attention to. For summaries of medical research and standards of care, 70% expect to utilize it within the year. For creation of discharge instructions, care plans, and/or progress notes, it is 65%.

Sam: And 64% expect to incorporate generation of chart summaries within the year. 61% say the same for documentation of billing codes, medical charts, or visit notes.

Maya: There is a really critical insight here regarding documentation. The report points out an enthusiasm gap. Between 57% and 58% are enthusiastic about the use case, although only 28% are currently incorporating it into their practice.

Sam: That is a 30-point gap between wanting a tool and actually having it in the workflow. If I am a software vendor, that tells me the demand for automated documentation is sky-high, but the current deployment or integration is lagging way behind what doctors actually want.

Maya: Exactly. And this ties directly into how physicians view the overall impact of AI on their practice. Sentiment is trending highly positive. Over three-quarters say AI provides an advantage in the ability to care for patients.

Sam: And that number has grown steadily. It was 65% in 2023, 69% in 2024, and now it is 76% in 2026. The skepticism is absolutely thawing.

Maya: They expect the greatest benefits in two very specific areas: work efficiency, where 78% expect it to be helpful, and diagnostic ability, where 74% expect it to help.

Sam: But it is not all sunshine. The report explicitly notes that a substantial share remain equally excited and concerned about AI. That is exactly 40% of the respondents.

Maya: Right. We have to look at what is driving that concern. Let us talk about burnout first, because that is often pitched as the main reason to buy AI. Seven in ten respondents see opportunities for AI to automate clinical tasks, and 73% anticipate automation of administrative duties.

Sam: Which sounds like a perfect recipe for reducing burnout. Yet, when asked how AI will impact stress and burnout, only 38% said it would be helpful, and 13% actually expect it to be harmful.

Maya: The concern about AI harming stress or burnout is low at 13%, but you are right, the helpfulness rating is not as universally high as you might expect given the automation numbers. However, the report states that these anticipated shifts reflect strong expectations that AI can relieve key drivers of burnout.

Sam: If burnout is not their primary fear, what is? The survey measured several factors, and there was only one area that emerged as a net-negative. Patient privacy.

Maya: This is a massive red flag for health-system IT buyers. 41% of physicians anticipate that tools using AI will be harmful to patient privacy. Only 7% think it will be helpful.

Sam: And the data shows exactly where that fear comes from. It is about who provides the tool. 71% are concerned about patient privacy when using non-institutional tools.

Maya: But for institutional tools, that concern drops to 42%. That is a huge gap. It means doctors trust their own hospital's vetting process significantly more than a standalone app they download on their phone.

Sam: Beyond privacy, there is a deep psychological fear highlighted in this document. Skill loss. 88% say they are very, somewhat, or mildly concerned about skill loss.

Maya: This is one of the most critical findings for the future of medicine. They are looking at the next generation of doctors. 70% are very or somewhat concerned about the loss of skills in current medical school students and residents.

Sam: If an AI drafts every progress note and suggests every differential diagnosis, how does a resident ever build the pattern recognition required to be an attending physician? That is the real worry here.

Maya: And it gets more interesting when you look at how doctors view their own skills. Overall, 28% are very or somewhat concerned about the loss of skills in themselves. But the breakdown by career stage is striking.

Sam: I have the numbers right here. For early career physicians, meaning those with less than 10 years in practice, 35% rank their concern for personal skill loss as a 4 or 5 out of 5.

Maya: Compare that to late career physicians, with over 30 years in practice. Only 21% of them share that high level of concern for personal skill loss. The younger doctors, who are still solidifying their clinical intuition, feel significantly more threatened by relying on automation.

Sam: There is also a specialty divide. Primary care physicians report 34% high concern for personal skill loss. For surgical specialties, it is 24%.

Maya: Which makes perfect clinical sense. Primary care relies heavily on synthesizing complex, often vague patient histories—exactly the kind of cognitive work AI is starting to target. Surgeons rely heavily on physical, procedural skills, which AI is not replacing anytime soon.

Sam: Quick note before we carry on. This spot is open for a sponsor. If your company builds or sells AI for healthcare and wants to reach the clinicians, health-system leaders and industry teams who listen to this show, the link to our sponsorship page is in the show notes.

Maya: And now, back to the document.

Sam: Let us shift from the doctors to the people sitting across from them. The patients. This survey has a fascinating section on physician views on patient AI usage. And there is a huge disclosure gap.

Maya: Massive. 29% of physicians report that they have not had a single patient disclose AI use to them. Zero.

Sam: And only 8% say a majority of their patients openly acknowledge using AI. So if you just look at what patients are saying out loud, you would think nobody is using it.

Maya: But the doctors know better. 30% of these physicians believe a majority of their patients are probably using AI for health questions. They are assuming a massive amount of unreported activity is happening in the background.

Sam: So how do the doctors actually feel about patients using these tools? Surprisingly, 70% say use of general-purpose AI chatbots for health information or decision-making is positive or has no impact.

Maya: They are okay with it, but only within strict bounds. The survey asked how much they want patients using AI for specific tasks. For questions about medications and medication side effects, 68% definitely or sometimes want patients using it.

Sam: And for general health questions, 64% say definitely or sometimes. But the moment the AI steps into diagnostic territory, the door slams shut.

Maya: Exactly. Nearly half of the respondents say they would never or rarely want patients using AI for interpretation of radiology or pathology results. Specifically, 49% for pathology and 46% for radiology reports and images.

Sam: They do not want a patient feeding a biopsy report into a chatbot at 2 AM. The anxiety and the potential for hallucinated bad news are just too high.

Maya: So we have talked about what doctors are using, and what they want patients doing. Let us look at what it actually takes to get an AI tool purchased and deployed. The accelerants and barriers.

Sam: If you are a vendor selling into a health system, this is the most important part of the document. The survey tested 21 different attributes for adopting AI. The top two requirements are non-negotiable.

Maya: 88% say it is important that its safety and efficacy is validated by a trusted entity and monitored over time. And 86% say it is important that data privacy is assured by my own practice/hospital and EHR vendor.

Sam: What does not matter as much? Recommendations from peers. The chart shows that only 46% rate other physicians I know are using it as important. They do not care if their buddy down the street uses it; they want hard proof it is safe.

Maya: Which brings us to the data spotlight in this report: liability. In cases where patient privacy, safety, or clinical efficacy are compromised, respondents want to know who holds legal responsibility and accountability.

Sam: The survey asked them to rank seven regulatory actions that would increase trust and adoption. Clear liability frameworks was most frequently selected as the most important. 31% selected it as rank number one. It had an average rank of 3.0.

Maya: This is the elephant in the room for medical AI. If a predictive algorithm misses early signs of sepsis, and the doctor follows the algorithm, who gets sued? Until clear liability frameworks are established, enterprise adoption is going to face friction.

Sam: The second highest regulatory priority, with an average rank of 3.4, was post-market surveillance, including adverse event reporting and technology audits. They want to know the tool is being watched after it is turned on.

Maya: And the third was increased oversight of AI-enabled medical devices by governing bodies such as the FDA, with an average rank of 3.6.

Sam: To build that trust, doctors want a seat at the table. 85% want to be consulted or be responsible for adoption of AI into their practice. You cannot just force this on them from the IT department.

Maya: When asked what level of evidence they look for, 50% want availability of peer-reviewed literature. 40% want U.S. clinical trials outlining the efficacy of the outcomes. They are holding AI to the exact same standard they hold a new pharmaceutical drug.

Sam: Which makes the training deficit we see in this report so glaring. 92% say they want more education and training on AI.

Maya: Right now, 73% of respondents have received some level of training. Which sounds okay until you realize that 27% have received no AI training from any source. And of those who did get training, only 11% report receiving a lot of training.

Sam: The breakdown of who is getting trained is totally counterintuitive. You would assume the younger, digital-native doctors are the most trained. But the data says the exact opposite.

Maya: It is a fascinating paradox. Only 55% of physicians under 35 say they have received any training. But for practicing physicians over 55, that number jumps to 71%.

Sam: Wait, older doctors are significantly more trained in AI than doctors under 35? How does the AMA explain that?

Maya: The report notes that training levels vary by practice setting, indicating that organizational support may depend on employer priorities and resource availability. It explicitly says a physician's age appears to play a smaller role in predicting exposure.

Sam: Ah, so the older doctors are likely entrenched in large hospital systems that mandate employer training, while the under 35 crowd might still be navigating the chaotic schedules of early career or residency without formalized employer tech rollouts.

Maya: Exactly. Look at the settings: 79% in a hospital setting have training, 76% in group practice, and 74% in solo practice.

Sam: So if 92% want more training, how do they actually want it delivered? They gave them a list to check all that apply.

Maya: They want it seamlessly integrated. 49% want it as part of my EHR training. 46% want it as I'm doing my job. 40% want it as part of an online continuing education course. And 39% asked for a live sandbox environment where I can experiment with tools.

Sam: What is dead last on the list? Getting trained by the AI companies themselves. Only 14% checked that box. They want to learn from their EHR vendor, their hospital, or their medical societies, not a vendor sales rep.

Maya: That is a huge takeaway for companies trying to drive adoption. You have to partner with the institutions doctors already trust.

Sam: Before we wrap up, we always have to talk about limitations. This is a robust survey of 1,692 physicians, but it is a survey.

Maya: Yes, the methodology section clearly states that results are based on self-reported data. More specifically, they note that adoption reflects physician self-reported incorporation of AI tools and does not independently verify organizational deployment. So a doctor might say they are using a tool, but that does not mean the hospital IT department formally bought it.

Sam: Which brings us right back to that shadow IT problem, both for doctors using personal tools and patients not disclosing their AI use. Maya, if a hospital chief medical officer is listening to this, what is the one memorable takeaway they need to bring to their next meeting?

Maya: It is that physicians are no longer debating whether AI belongs in medicine. They are already using it, averaging 2.3 use cases. The conversation has completely shifted from 'Does this work?' to 'Who gets sued if it fails, and how do we prevent our residents from losing their clinical skills?' You have to solve for liability and skill loss, not just efficiency.

Sam: A perfect summary. The 2026 AI winter is officially dead in the exam room.

Maya: To recap, over 80% of surveyed physicians now use AI professionally, with strong excitement for documentation and research summaries. But adoption hinges on clear liability frameworks and hard proof of safety, amidst growing concerns over how this tech impacts trainee skill development.

Sam: You can find a link to the full AMA 2026 Physician Survey on Augmented Intelligence in our show notes. And as always, this podcast is for informational purposes only and is not medical advice.