Physician AI use is surging. Is your practice ready?

Physician adoption of artificial intelligence has more than doubled in three years, with clinicians increasingly turning to the technology for documentation support and clinical decision-making. But the rapid uptake is outpacing institutional readiness, leaving practice administrators to close a widening gap between what their staff is already using and what the organization has formally approved, trained for or written policy around.

That is the central tension in new survey data from the American Medical Association’s Center for Digital Health and AI, released March 12. AMA CEO John Whyte, M.D., M.P.H., said in a statement that AI has quickly become part of everyday medical practice and that physicians see real promise in its ability to support clinical decisions and reduce administrative burden. But he cautioned that the tools must be safe, effective and designed to enhance, not replace, physicians.

The AMA’s 2026 Physician Survey on Augmented Intelligence, conducted annually since 2023, found that 81% of physicians now use AI in their practices, up from 38% three years ago. The average number of use cases per physician has climbed to 2.3 from 1.1 over that span. The most common applications center on medical research summarization and clinical care documentation, two areas where administrative staff and practice leaders are often the ones managing the downstream workflow.

Confidence is rising alongside adoption. More than three-quarters of physicians now believe AI improves their ability to care for patients, compared with 65% in 2023. Diagnostic accuracy and work efficiency ranked as the greatest expected advantages.

About 70% of physicians view AI as a tool to automate tasks that contribute to work-related burnout, a persistent concern for administrators trying to retain staff. But the optimism comes with caveats: 88% expressed concern about potential skill loss, especially among those with 10 years or less in practice, and 40% described their outlook as equally excited and concerned, citing patient privacy and the integrity of the patient-physician relationship as top worries.

A workforce-wide trend, with an administrator-shaped hole in the middle

The AMA data is not an outlier. A separate 2025 survey from Offcall, which polled more than 1,000 physicians across 106 specialties, found that 67% were using AI daily and nearly 90% used it at least weekly. That same survey found that 81% were frustrated with how their organizations handle AI deployment, and 71% said they had little to no influence over which tools get selected or implemented. For administrators, that frustration is a signal: clinicians want to be consulted, and they want leadership to move faster.

Among physician assistants, the picture is strikingly similar. A Wolters Kluwer survey published in late 2025 found that 56% of PAs reported using AI daily and that clinical documentation and ambient scribe technologies were the most common applications. But 87% acknowledged needing more AI training, 83% wanted more formal employer-led instruction, and only 32% said their workplace had clear guidelines for safe use.

Kelly Villella, director of medical education and practice at Wolters Kluwer Health, told Physicians Practice that the gap between adoption and preparation is not unique to PAs and carries real consequences for administrators overseeing clinical staff.

“Although their use of AI is growing, 87% still feel like they need more AI training, and I think that’s true across a lot of health care and other professions,” Villella said. “AI is changing so rapidly, people are trying to keep up.”

Villella said the survey revealed strong confidence among PAs when it comes to patient interaction, with 96% feeling prepared for face-to-face care. But documentation stood out as a persistent friction point, particularly for administrators responsible for onboarding new clinicians into practice-specific systems.

“Documentation is a challenge. If you think about it, there’s preparation, but then each individual practice has different systems, so always in onboarding, you have to onboard in the systems,” she said. “And possibly even think about what are best practices so that they can focus on the patient while doing all the necessary administrative paperwork.”

She described the broader source of workflow friction in blunt terms.

“That’s the friction. It’s a necessary workflow, but in some ways, it takes you away from the passion and the mission,” Villella said. “How can we make it more efficient to keep that documentation? I think that’s the opportunity for AI as well, to integrate into the workflow in ways that are protected, secure.”

For practice leaders weighing how to balance productivity expectations with quality, Villella said the answer starts with thinking about AI as a feature within existing tools, not a standalone solution.

“Instead of a be-all, end-all, like AI is a solution, AI is a feature that is augmenting a solution and making it even more powerful for you in your everyday practice, so you can be more efficient,” she said. “Practices should be thinking about how they are starting to evaluate even the current solutions that they have to make sure they have this expert AI that’s trusted, built into those solutions.”

She also issued a pointed warning about shadow AI, the use of unapproved AI tools by staff members who bring personal habits from consumer technology into clinical settings.

“There are all kinds of open sources proliferating, and you might think, because you use it day to day now in your personal life, as a new clinician, ‘Oh, let me run my notes through this,’” Villella said. “I think people don’t even all know the term shadow AI. There’s acceptable AI, and things that you do in your day to day you might not be able to leverage when you’re in the practice.”

Villella urged practice managers to set explicit expectations with their staff about the boundaries of AI use, including when ambient listening tools and transcription are appropriate and what kind of human review is required afterward.

“If you’re going to leverage AI, let’s say, to transcribe, read it through again, make sure there isn’t a mistake,” she said. “It’s not doing your whole documentation job for you. It’s an aid. So getting really explicit about that, what the expectations are, you can leverage these tools, but then you need to have checks and balances.”

Governance, liability and the regulatory picture

The AMA survey found that physicians overwhelmingly want a say in how AI tools are adopted, a finding that should resonate with administrators making purchasing and policy decisions. Eighty-five percent of physicians said they want to be consulted or directly involved in adoption decisions. They also rated data privacy (86%) and robust safety and efficacy validation (88%) as critical preconditions for broader adoption, and clear liability frameworks topped the list of regulatory actions most essential to building trust.

The governance question extends beyond the clinical setting and into a policy environment that is adding new layers of complexity for practice leaders. Anders Gilberg, senior vice president of government affairs at MGMA, told Physicians Practice that cybersecurity regulation is shaping up as a significant cost and compliance issue for medical groups in 2026.

“This administration keeps talking about a binary rule related to cybersecurity, and our concern is an unfunded mandate, potentially requiring medical groups to spend extremely high dollars if they finalize what was proposed under the Biden administration,” Gilberg said.

Gilberg said the regulatory landscape extends well beyond cybersecurity. He noted that practices are simultaneously navigating short-term telehealth extensions, potential lab reimbursement cuts, the expiration of ACA subsidies and uncertainty around value-based care incentives. For administrators, those pressures compound the challenge of also building an internal AI governance framework.

“Practices have to change midstream, only to then have Congress reinstate policies a few weeks later,” Gilberg said, describing the whiplash effect of short-term legislative fixes on practice operations. “When Congress can’t agree, or delays implementation, practices feel it immediately.”

Gilberg also pointed to coverage disruptions as a looming concern for practice revenue. With ACA premium tax credits expired, he warned that practices whose payer mix includes exchange coverage may see patients unable to renew.

“That can mean payment plans, discounts for cash-pay arrangements and other accommodations, but practices also have to protect their bottom line,” Gilberg said. He added that coverage shocks raise the risk that patients skip primary care visits entirely and end up in the hospital with conditions that were once manageable.

“Practices have to get much more aggressive about eligibility verification, even for existing patients, because coverage can change or lapse at year-end,” he said.

Patients are using AI, too, and practices need a plan for that

The AMA survey also surfaced tension around how patients engage with AI. Physicians generally support patient use of the technology for general health and medication questions, but nearly half strongly oppose patients using AI to interpret radiology or pathology results, tasks that require clinical judgment.

Amber Maraccini, Ph.D., M.A., vice president and head of health care at Medallia, told Medical Economics in a recent interview that patients arriving with AI-generated conclusions is not fundamentally different from patients who used to bring in printouts from Google searches. The productive approach, she said, is for clinicians to avoid dismissing the effort and instead invite the patient to review the information together.

For administrators, that means front-desk workflows and visit-prep protocols may need updating. When patients show up with AI-sourced questions or conclusions, the clinical team needs a consistent approach, something that flows from a practice-wide understanding rather than individual clinician judgment in the moment.

The education pipeline matters for hiring managers

Villella emphasized that the training gap is not just a practice-level issue. It reaches back to the education pipeline, which matters to administrators thinking about what new hires will arrive prepared to do.

“Practice begins with education,” Villella said. “I think there’s a great message in the survey for the PA programs that are educating to also think about how they are starting to model use of AI now in education, how they are starting to communicate acceptable use or not in the curriculum. So that when a practice manager is getting a new clinician, they already have a mindset related to this.”

Until the education side catches up, Villella said, the burden falls on practice leadership.

“When they get to practice, they can be trained on the specific tools, the specific use cases, but they still have a mindset towards how to think about it and appropriate use, because it is going to be transformative,” she said.

The bottom line for practice leaders

The data paints a consistent picture across multiple surveys and clinician types: AI adoption is widespread, accelerating and largely outrunning institutional support. For administrators, this is no longer a question of whether to engage with AI but how to bring structure to something that is already happening inside their walls.

“There’s a lot happening in health care, a lot of technology programs moving forward,” Gilberg said. “I think it’s going to be an interesting year in 2026.”

Villella put it more directly when asked what practice managers should do next week.

“Sit down as a team and work on writing policies related to acceptable use, safe use,” she said. “Don’t assume, because it’s so nascent and evolving, that every clinician coming in has the same understanding. It’s best at this point, where things are changing so much, not to assume anything and just sit down and document your current policy.”

Five things practice administrators can do now

1. Draft a written AI acceptable-use policy. Villella said this is the single most actionable step a practice can take. The policy should clearly define which AI tools are approved, how and when they can be used, and what is off-limits. It should also address shadow AI directly, including personal subscriptions to consumer tools like ChatGPT that clinicians may be using on their own. The policy does not need to be long, but it needs to exist, be communicated during onboarding and be revisited regularly as the technology evolves.

2. Audit the AI capabilities already in your existing systems. Many EHR vendors and clinical decision-support platforms have added AI-powered features in recent updates. Before shopping for new tools, Villella recommended that administrators evaluate what is already available in their current vendor solutions and determine whether staff are aware of those features and trained to use them. This can surface quick wins without adding new contracts or compliance risk.

3. Build AI training into onboarding and ongoing education. The Wolters Kluwer survey found that 83% of PAs want more formal, employer-led AI training, and the AMA data shows that training and education rank among the top facilitators of adoption. Administrators should include AI orientation in their onboarding process, covering approved tools, documentation expectations and the human review required when AI-generated content is involved. For existing staff, periodic refreshers will be necessary as tools and policies change.

4. Involve clinicians in AI purchasing and implementation decisions. The AMA found that 85% of physicians want to be consulted or directly involved in AI adoption decisions, and the Offcall survey found that 67% said having more influence would improve their job satisfaction. Administrators who include clinical staff in vendor evaluations, pilot programs and policy development will not only make better purchasing decisions but also build the buy-in that drives successful implementation.

5. Prepare your front office for patients who bring AI-generated information to visits. Nearly half of physicians in the AMA survey strongly oppose patients using AI to interpret radiology or pathology results. As more patients arrive with AI-sourced conclusions, practices need a consistent framework for how clinicians and support staff handle those conversations. That starts with a simple script: acknowledge the patient’s effort, review the information together and redirect to clinical context. Training front-desk and intake staff to flag these situations can help ensure visits stay productive.