AI Is Already in the Healthcare Workflow. Trust Is Still Catching Up.

The healthcare conversation about AI can make it sound as if its place in the workflow is still up for debate.

It isn’t.

Healthcare professionals are already using AI to draft documentation, summarize research, and, in some cases, support clinical decisions. Some of that use is formally approved and tracked. Some is happening less visibly, one login at a time, while organizational policies and training work to catch up.

AI has entered the chat and, in some cases, started drafting the notes. Healthcare professionals are also using it to summarize research, support administrative work, and, in some settings, assist with clinical decisions. Some of that use is formally approved and integrated. Some of it is individual, informal, and much harder for an organization to see.

That doesn’t mean trust has arrived at the same pace. It means healthcare now has to manage two realities at once: AI can be useful enough to become routine while remaining uncertain enough to require scrutiny.

Adoption Doesn’t Mean Confidence

Healio’s May 2026 survey of 618 healthcare professionals across more than 35 specialties captures this tension clearly. About 72% of respondents said they use AI professionally either regularly or occasionally, yet only 18% reported strong trust in AI for clinical decision-making.¹

The contrast becomes more apparent when AI moves closer to the patient. Nearly seven in ten respondents were comfortable using it behind the scenes, while only 36% were comfortable using it directly in front of patients. More than 40% of respondents whose organizations didn’t support AI use said they were nevertheless using AI tools professionally.

These numbers aren’t necessarily contradictory. A clinician may use AI to create a first draft, search a large body of information, or reduce repetitive work while still checking the result carefully. Use can signal convenience or utility without representing broad confidence in the technology.

A larger 2026 American Medical Association survey points in the same direction. More than 80% of the 1,692 physician respondents reported using AI professionally, while 40% said they felt equally excited and concerned about its impact.² Adoption and caution are not opposites here. Healthcare professionals can use AI and still have reasonable concerns about it.

Written Policy Is Falling Behind Actual Practice

For healthcare leaders, the practical problem isn’t simply whether an organization has an AI policy. It’s whether that policy reflects how people are already working.

Healio found that only 16% of respondents reported clear organizational AI policies, and just 11% had received structured institutional training.¹ When daily behavior advances faster than formal guidance, an organization can end up governing the AI it has approved while missing the AI its workforce has actually adopted.

A useful starting point is visibility.

• • Which tools are people using? For which tasks?

• • What information is being entered?

• • Where are outputs reviewed by a human?

• • When does AI become visible to patients?

• • How can someone report an error or questionable recommendation?

Those questions are less dramatic than a sweeping debate about whether AI is good or bad. They’re also more actionable.

Some AI Use Cases Carry More Risk Than Others

The level of caution should follow the consequence of the task. Using AI to reorganize internal notes isn’t the same as relying on it to interpret an image, suggest a diagnosis, or communicate directly with a patient. Treating every use case as equally risky can make guidance too broad to be useful. Treating every use case as equally promising can understate meaningful clinical and operational differences.

Healio’s respondents reflected that distinction. Their most common uses centered on documentation, decision support, and research. They were also far more comfortable with AI behind the scenes than in patient-facing moments. The technology may be entering healthcare broadly, but confidence remains highly dependent on context.

Patient research adds another layer. A 2026 JAMA Network Open study of 3,000 U.S. adults found that AI performance had the strongest association with trust and choice.³ Clinician presence, representative data, and formal governance also mattered. To put it simply, patients weren’t responding to the idea of AI alone. They were responding to evidence about how well it worked, who remained involved, and what safeguards surrounded it.

Trust Isn’t a Communications Deliverable

Clear communication matters, especially when AI touches the patient experience. But communication can’t carry more weight than the underlying evidence and operating practices.

The Coalition for Health AI reported in January 2026 that 75% of surveyed adults used AI, while only 13% felt very comfortable with it.⁴ More than 80% said clear accountability measures would increase their trust. That finding is a useful reminder: disclosure may tell people that AI is present, but accountability helps explain why its presence should be accepted.

Organizations should be prepared to answer practical questions in clear language.

• • What does the tool do?

• • What does it not do?

• • How was it evaluated?

• • Who reviews its work?

• • What happens when it is wrong?

Those answers should be consistent across leadership, clinical teams, training, patient communication, and external messaging.

Clinicians Want a Role, Not Just a Rollout

One of the more consistent signals across the research is the desire for meaningful clinician involvement. In the AMA’s 2026 survey, 85% of physicians wanted to be consulted about or responsible for AI adoption in their practice.² Ninety-two percent wanted more education and training.

That preference matters because trust is partly built through participation. Inviting clinicians into evaluation, workflow design, implementation, and feedback can surface concerns that may not appear in a product demonstration. It can also help distinguish between resistance to change and reasonable questions about accuracy, privacy, liability, or fit.

Training should be equally practical. A general overview of AI may build awareness, but teams also need guidance tied to specific tools and decisions: what data can be entered, which outputs require verification, how performance is monitored, and where to go when something doesn’t look right.

A More Useful Place to Begin

Healthcare organizations don’t need to manufacture certainty where it doesn’t yet exist. They do need to recognize that waiting for universal confidence will not pause everyday use. 

The near-term work is more concrete:

• • Understand current behavior

• • Separate lower-consequence support tasks from higher-consequence clinical applications

• • Involve healthcare professionals in decisions

• • Define appropriate review

• • Create training that fits the workflow

The gap between use and trust won’t close through enthusiasm alone, nor should trust be rushed. In healthcare, confidence should be earned through evidence, experience, transparency, and accountability. 

There’s a modestly encouraging signal in the data. Healthcare professionals aren’t uniformly closing the door on AI. They’re describing the conditions where it can become more useful, more responsible, and more worthy of confidence. 

Trust can’t be rushed. Readiness can be built. 

Healthcare professionals are already using AI. The work now is to build practical guardrails around how it’s being used.