6 November 2026 · Medows
91% of Patients Want to Approve Every AI Use
A Tebra survey finds 91% of patients want permission every time AI supports diagnosis. Yet 9% of practices run AI with no review or policy.
A new survey puts a number on something every ward doctor already feels. Patients want to be asked before AI touches their care. Not once, at registration. Every time.
Tebra, a practice-management company, published its AI in Private Practice report in October 2026. It surveyed 401 US adults and 292 providers, practice owners and practice leaders who currently use or pilot AI. The headline from the patient side: 91% want their permission required every time AI is used to support diagnosis or treatment.
What the survey actually found
The consent number is the loud one. The quieter numbers around it are more useful.
On the patient side, 49% said they are uncomfortable with AI in diagnosis, and 49% with AI in treatment. Only 13% were uncomfortable with AI in scheduling. The gap by age is striking: 67% of Gen Z respondents were uncomfortable with AI diagnosis, against 36% of Baby Boomers.
So patients are not anti-AI. They are anti-surprise. The closer the tool gets to a clinical judgment, the more they want a say.
On the provider side, 95% have AI live in at least one workflow and 60% use it daily. But only 7% have reached advanced or fully integrated use. The most common uses are ambient documentation (74%), patient messaging (65%), scheduling (63%) and phone or intake (54%).
Respondents reported documentation time falling from about 15 to 10 hours a week. Yet only 48% reported burnout relief. Saving the hours did not automatically make the job feel lighter.
The oversight gap
This is the part that should make a ward team pause. In the same survey, 9% of providers run AI with no accuracy review, no patient disclosure and no compliance policy. Another 27% have no formal compliance policy or staff training. The top barrier named by providers was accuracy concerns, at 32%.
Put those side by side. Patients say: ask me every time. A meaningful share of practices say: we have no written policy on it.
Think of the scene. A registrar is on a busy round. A tool drafts a summary. Nobody on the team is sure whether the patient was told it was used, or who checked the output. The tool is helpful. The process around it is a blank.
Read it with care
The authors are upfront about the limits. Both samples came from opt-in online panels (CloudResearch Connect and Prolific). They are nonprobability samples, so the authors call the results directional and say they cannot be projected to all practices or all US patients. Margins of error are roughly 5.7 points for providers and 4.9 points for patients.
Also note the setting. This is a private practice report, mostly about outpatient workflows. A hospital ward is a different place. Still, the direction of the patient signal is hard to ignore.
What this means for doctors on rounds
Three things stand out to us.
First, consent is a workflow problem, not only a legal one. If asking every time is the expectation, it has to be easy and visible. A line buried in a form will not do it.
Second, accuracy review has to belong to someone. When a tool's output goes into a patient's record, the doctor signing it is the final check. That only works if the output is easy to verify against its source.
Third, hours saved is not the same as a better shift. The burnout number is a reminder that tools which bolt onto one task do not fix the whole day.
This is where we think the workspace idea matters. A tool that shows its sources, keeps the doctor in the loop and sits where the work already happens makes disclosure and review a natural part of the round. A point tool that disappears into the background makes both harder.
None of this is clinical guidance. It is a read on where patient expectations are heading, and how far some practices still have to go.
Sources
- Tebra, AI in Private Practice report: https://www.tebra.com/theintake/healthcare-reports/ai-in-private-practice-report
- Health IT Today, Bonus Features, November 1, 2026: https://www.healthcareittoday.com/2026/11/01/bonus-features-november-1-2026-91-of-patients-want-their-permission-required-when-ai-is-used-for-diagnosis-64-of-providers-plan-to-deploy-ai-for-claims-and-denials-management-pl/
Medows is a clinical AI workspace for the doctor on rounds. Learn more or write to alapan@medows.ai / alapanx@gmail.com.