Estimated reading time: 7 minutes · Last updated:
Patient comfort with AI in clinical care depends on whether a clinician keeps the final say, and primary care doctors seem to underrate it. In a Gallup Panel survey of 5,330 U.S. adults taken 12-24 August 2026, 56% said they would accept AI deciding on a diagnosis or treatment if their provider could change or overrule the call. Just 9% accepted those decisions with no provider review. A companion poll of 1,007 primary care physicians put the share of doctors expecting most patients to accept the override model 20 percentage points lower. The West Health-Gallup Center on Healthcare in America released the work on 9 November 2026. It also shows 89% of those physicians use AI at work, while 20% have formal training.
Key takeaways
- Override wins: In Gallup's survey, 56% of U.S. patients accept AI diagnosis or treatment decisions a provider can overrule, against 38% when AI only assists and 9% when nobody reviews it.
- Doctors guess low: Primary care physicians' estimates of patient acceptance for override-style AI fall 20 points short on diagnosis, 11 on imaging review and 7 on medication decisions.
- Review beats price: Among patients wary of autonomous AI, the right to request a provider review swayed just under half, while lower costs, faster decisions, privacy protections and safety evidence each swayed no more than a third.
- Use outpaces training: Of 1,007 surveyed primary care physicians, 89% used AI at work in the past year, yet 20% report formal training, dropping to 11% in independent practices.
Table of contents
- Key takeaways
- Why Patients Prefer AI Decisions a Clinician Can Overrule
- Even Patients Who Trust AI Want a Doctor Watching
- Where Primary Care Doctors Misread Their Patients
- Doctors Already Use AI Widely With Little Formal Training
- What favours and what limits wider clinical AI use
- What to be careful about
- Frequently asked questions
Why Patients Prefer AI Decisions a Clinician Can Overrule
Gallup tested three clinical jobs for AI: diagnosis and treatment, reviewing medical images, and choosing medications. For each one, patients rated three set-ups. In the first, AI offers input and the provider makes the call. In the second, AI decides and the provider may overrule it. In the third, AI decides and no clinician checks the result.
The override arrangement won every time. For diagnosis and treatment it drew 56% comfort, well ahead of 38% for AI as an assistant and 9% for full autonomy. Imaging review scored 53% for override, 52% for assistance and 13% for autonomy. Medication decisions came in at 49%, 45% and 11%.
Diagnosis is the odd case. Patients were about 18 points more at ease letting software make the first call, with a doctor holding a veto, than letting the doctor decide with software advice. The survey does not say why. Patients may see the override model as a doctor double-checking a machine, but that is a reading, not a finding. What the data shows plainly is that removing the clinician collapses support in all three uses.
Even Patients Who Trust AI Want a Doctor Watching
Support for machine-only decisions stays low among people already warm to the technology. Some 23% of patients say they trust the accuracy of AI-generated health information, 21% somewhat and 2% strongly. Across the three clinical uses, between 65% and 71% of that group accept AI that assists a provider or decides subject to a provider's veto. Without a provider, acceptance falls to between 18% and 20%.
The same split appears among the 39% of patients who used AI in the past year to look up physical or mental health information or advice. They showed 51% to 66% comfort when a provider stays involved in either form, and 12% to 16% when nobody reviews the AI.
Gallup then asked patients who rejected autonomous AI for diagnosis or imaging what might change their minds, offering five options. The right to ask for a provider's review came closest to a majority, chosen by just under half in both scenarios. Strong privacy protections, evidence that the AI is safe and effective, lower healthcare costs and quicker decisions each moved no more than a third. Cheaper, faster care does not buy acceptance; a route back to a human does.
Where Primary Care Doctors Misread Their Patients
Gallup ran a parallel opt-in survey of 1,007 licensed, practicing primary care physicians holding MD or DO degrees, fielded 7-23 August 2026. Each was asked whether most of their patients would be comfortable with each AI set-up. The doctors correctly expected autonomous AI to be the least welcome.
They missed on the override model. For diagnosis and treatment, the share of physicians who thought most patients would accept it ran 20 percentage points below the 56% of patients who did, placing the doctors near 36%. The gap narrowed for imaging review, 53% of patients against 42% of physicians, and for medication decisions, 49% against 42%. On assistive AI, physician estimates landed within two points of patients for diagnosis and medications but trailed by 11 points on imaging.
The West Health-Gallup Center argues that misjudging patients could mean missed chances to adopt AI tools they would welcome. Two cautions apply. The physician sample was opt-in, not probability-based. And the two groups answered different questions: patients gave their own comfort, while doctors judged whether a majority of their patients would agree. The size of the gap should be read loosely.
Doctors Already Use AI Widely With Little Formal Training
Physicians are not waiting for patients. Of the 1,007 surveyed, 89% used an AI tool or chatbot in the past year for at least one of seven work tasks, and 66% did so weekly or more often. Looking up medication information (78%) and researching symptoms or conditions (76%) led the list.
Use reaches into patient records too. Some 53% have used AI to review patient-specific material such as test results, scans or imaging, and 50% to draft clinical notes. Writing patient instructions or after-visit summaries stood at 49%, searching or summarizing medical records at 47%, and recording and transcribing appointments at 45%.
Formal preparation lags far behind. Just 20% of these physicians report any formal training in using AI in healthcare settings: 30% in hospital or health system practices, 16% in physician groups and 11% in independent practices. The Center frames the open issue as which kinds of training, guidance or support would help providers use AI responsibly when it shapes clinical decisions. That matters because patients' preferred safeguard assumes the clinician knows when to overrule the machine.
| Clinical use | AI assists, provider decides | AI decides, provider can override | AI decides, no review | Physicians expecting most patients to accept override |
|---|---|---|---|---|
| Diagnosis and treatment | 38% | 56% | 9% | About 36% |
| Imaging review | 52% | 53% | 13% | 42% |
| Medication decisions | 45% | 49% | 11% | 42% |
What favours and what limits wider clinical AI use
The case for
- Patients show majority or near-majority comfort with override-style AI: 56% for diagnosis, 53% for imaging and 49% for medication decisions.
- Patients who used AI for health information in the past year show 51% to 66% comfort when a provider stays involved.
- Physician estimates already match patients within two points for assistive AI in diagnosis and medication decisions.
The case against
- Autonomous AI draws only 9% to 13% patient comfort, a steep barrier for any tool that removes clinician review.
- With 89% of physicians using AI and 20% formally trained, the override safeguard patients favour rests mostly on clinicians without formal AI training.
- The physician survey was opt-in with a ±3.3 point credibility interval, so the size of the perception gap is less certain than the patient figures.
What to be careful about
- The provider findings come from a non-probability opt-in panel, unlike the probability-based patient sample with its ±1.7 point margin.
- Patients reported their own comfort while physicians judged whether most of their patients would be comfortable, so the two measures do not line up exactly.
- Both surveys ran online in English, leaving out anyone without internet access.
- Imaging and medication results rest on random subsamples of 2,653 and 2,662 patients rather than the full 5,330.
The bottom line
Gallup's paired surveys point to a design lesson rather than a verdict on AI in medicine. Patients accept software in diagnosis, imaging and prescribing on one condition: a clinician who can step in. Strip that out and comfort sinks to between 9% and 13%, and it stays at 20% or below even among patients who trust AI. Primary care doctors grasp that second point but understate the first, by 20 points on diagnosis. Meanwhile 89% of them already use AI at work and 20% report formal training. The next useful evidence will show whether health systems build override rights into their tools and whether training reaches independent practices, where it stands at 11%.
What to watch
- Watch for the next West Health-Gallup release revisiting patient and physician views on clinical AI; the Center says tracking change over time matters, but no date has been set.
- Watch whether hospital systems, physician groups and independent practices raise formal AI training above 30%, 16% and 11%; no follow-up survey has been scheduled.
Frequently asked questions
How reliable is Gallup's survey on patient comfort with AI?
The patient survey used the probability-based Gallup Panel and carries a margin of sampling error of ±1.7 percentage points at 95% confidence. The physician survey was a non-probability opt-in panel with a credibility interval of ±3.3 points. Gallup weighted patients to 2024 American Community Survey estimates and physicians to the AHRF/AMA Physician Masterfile.
Did every patient answer every AI question in the survey?
No. Gallup tested 13 possible AI applications and each patient rated seven of them. All 5,330 answered on diagnosis and treatment, while imaging review went to a random subsample of 2,653 and medication decisions to 2,662. Each of the 1,007 physicians rated all 13.
How does this fit with earlier West Health-Gallup research on AI and health?
Earlier West Health-Gallup research found Americans who use AI for health information mostly do so alongside regular care, though some skipped a provider visit after getting AI advice. The West Health-Gallup Center on Healthcare in America says this study is the first time it has surveyed patients and providers side by side on AI.
Related reading
