Estimated reading time: 6 minutes · Last updated:
Most American patients are open to AI in healthcare helping with diagnosis and treatment decisions, as long as a doctor can overrule it. Their doctors underestimate that openness. Surveys by the West Health-Gallup Center on Healthcare in America found that 56% of patients are comfortable with AI-driven calls when a physician can override them. Among clinicians, only 36% believed most of their patients would feel that way, a gap of 20 percentage points. Patient acceptance has a firm limit: just 9% would let AI decide with no doctor reviewing the result. Clinicians are already using the technology. 89% said they had used AI at work in the past year, but only 20% reported any formal training in it.
AI has enormous potential to improve healthcare, but patients are sending a clear message that they are not afraid of using the technology, and want their providers involved, not replaced.
Timothy Lash, president and chair of the West Health Policy Center
Key takeaways
- The expectation gap: In the West Health-Gallup surveys, 56% of patients accepted AI diagnosis or treatment decisions that a doctor could override, but only 36% of clinicians expected most patients to accept them.
- Oversight is the condition: Only 9% of the 5,330 U.S. adults surveyed wanted AI to make medical decisions without a doctor's review.
- Comfort collapses without a doctor: Patient comfort with AI reading scans fell from 53% with a doctor's override to 13% with no doctor involved. For medication choices it fell from 49% to 11%.
- Use outpaces training: Of 1,007 clinicians surveyed, 89% had used AI for work in the past year, while 20% had received any formal AI training.
- Primary care relies on it: Primary care doctors most often use AI to look up medication information (78%) and symptoms and conditions (76%), and 53% have used it on test results or imaging.
Table of contents
- Key takeaways
- How Wide Is the Gap Between Patients and Doctors on AI?
- Why a Doctor's Override Is the Price of Patient Approval
- Clinicians Use AI Widely but Few Have Been Trained
- What Primary Care Doctors Already Hand to AI
- What could help or slow AI in clinical decisions
- What to be careful about
- Frequently asked questions
How Wide Is the Gap Between Patients and Doctors on AI?
The West Health-Gallup Center on Healthcare in America ran two parallel surveys. The first reached 5,330 U.S. adults, and its findings carry a 1.7-percentage-point margin of error. The second reached 1,007 clinicians, with results accurate to within 3.3 percentage points. The West Health Institute released the findings in a news release dated 10 November 2026.
The central finding is that doctors and patients expect different things. Among patients, 56% said they would accept AI reaching diagnosis or treatment decisions if their own doctor could amend or reverse them. Among clinicians, just 36% thought most of their patients would accept that arrangement.
That 20-point difference has practical weight. A clinician who plans around patient resistance that the data does not show could misjudge how to introduce these tools at the bedside. Joe Daly, Gallup Global Managing Partner, read the results as encouraging. In a news release he said: "These findings suggest reason for optimism about the role of AI in healthcare."
Why a Doctor's Override Is the Price of Patient Approval
Patient comfort comes with a condition attached. Only 9% wanted AI to settle medical questions with no physician checking the outcome. Set against the 56% who accept AI when an override exists, that figure shows patients are endorsing a supervised tool, not an automated clinician.
The pattern holds when the task is named. When asked about AI examining scans such as X-rays or MRIs, 53% were comfortable provided a doctor could overrule the result. 52% were comfortable with a doctor doing the reading with some AI help. For prescriptions chosen on the basis of a patient's history, the equivalent figures were 49% and 45%. With the doctor removed entirely, comfort dropped to 13% for imaging and 11% for medication. That is a fall of roughly 40 points and roughly 38 points respectively.
Daly said this boundary is the point of the research. "The data show that patients are open to AI having a role in their care, particularly when they know a clinician can still step in," he said. He added that as AI spreads through medicine, it will be critical to map which uses patients accept and where they still want a person in charge.
Clinicians Use AI Widely but Few Have Been Trained
The provider survey shows that AI is already part of clinical work. Nearly 9 in 10 respondents, 89%, said they had used it in their jobs during the past year. Only 1 in 5, or 20%, said they had received formal instruction of any kind in how to use it.
The distance between those two figures is about 69 points. Many clinicians using these tools have learned them on their own, or not at all. The survey results do not say what counted as formal training or who provided it.
Timothy Lash, president and chair of the West Health Policy Center in Washington, D.C., linked the two surveys. He argued that patients are comfortable with the technology but want their doctors kept in the loop rather than swapped out. In his view, that raises the stakes for clinician training. "This makes it critical that providers have the training to use it effectively and patients have trust in how AI is being used to guide medical decisions," Lash said in a news release.
What Primary Care Doctors Already Hand to AI
Primary care doctors described AI as part of their routine. Their most frequent use was looking up information. 78% have used it to research medications and 76% to research symptoms and conditions.
Clinical tasks come next. Some 53%, more than half, say they have turned to AI when reviewing imaging or test results, while drafting clinical notes with it is something 50% have done. The imaging figure lines up with the patient side. An identical 53% of patients said they would accept AI reading their scans as long as a doctor could overrule it.
Paperwork and record-keeping uses fall just under half. AI has helped 49% prepare patient instructions or after-visit summaries, and 47% pull or condense information from medical records. 45% have used it to record and transcribe appointments. In each of these uses the doctor still makes the decision, which is the arrangement patients in the companion survey said they wanted.
| Arrangement | Reviewing medical images | Medication decisions |
|---|---|---|
| AI decides, doctor can override | 53% | 49% |
| Doctor decides with some AI assistance | 52% | 45% |
| AI alone, no doctor involved | 13% | 11% |
What could help or slow AI in clinical decisions
The case for
- 56% of patients already accept AI-driven diagnosis or treatment calls when a doctor can override them, more than the 36% of clinicians expected.
- 89% of clinicians report using AI at work, so supervised tools are already in daily use.
- Patient comfort with AI-reviewed scans (53%) is almost the same as with doctor-led review assisted by AI (52%), so both models have similar patient support.
The case against
- Only 20% of clinicians report formal AI training, which Timothy Lash says is critical for effective use.
- Patient support falls to 13% for imaging and 11% for medication once doctors are removed, which limits any move toward unsupervised AI.
- Clinicians who underestimate patient acceptance by 20 points may be slow to discuss AI use openly with patients.
What to be careful about
- Clinicians are using AI far more widely (89%) than they are being trained to use it (20%), so tools may be applied without structured guidance.
- The surveys measure stated comfort, not how patients react when AI is actually used in their own care.
- The clinician survey's 3.3-point margin of error is wider than the patient survey's 1.7 points, so the provider figures are less precise.
- Patient acceptance depends on a doctor's ability to override AI, and the surveys do not show how often that override is used in practice.
The bottom line
The West Health-Gallup results show a public that is more open to AI than its doctors assume, but only on one condition. Patients will accept AI in diagnosis, imaging and prescribing when a physician keeps the final say. Their support drops to 13% or less once that safeguard is gone. Clinicians, meanwhile, are using AI far more than they are being taught to use it: 89% report using it and 20% report formal training. The open question is whether health systems close that training gap before the tools take on more clinical work. Lash argues that patient trust depends on closing it, and nothing in the data suggests the step can be skipped.
What to watch
- Watch for any further breakdown of these results from the West Health-Gallup Center on Healthcare in America, for example by age or medical specialty; no date has been given.
- Watch whether the 20% formal-training figure among clinicians changes in any repeat of the provider survey; no date for a follow-up has been set.
Frequently asked questions
How reliable is the West Health-Gallup survey on AI in healthcare?
On the patient side, 5,330 U.S. adults took part, and their answers could be off by 1.7 percentage points either way. On the clinician side, 1,007 providers responded, with a 3.3-point margin. Either margin is smaller than the 20-point gap between how comfortable patients are and what doctors expect.
Who conducted the survey on patient attitudes to AI medical decisions?
The West Health-Gallup Center on Healthcare in America ran both surveys, and the West Health Institute released the results on 10 November 2026. Gallup's Joe Daly and the West Health Policy Center's Timothy Lash commented on the findings.
Are patients more comfortable with AI reading scans or choosing prescriptions?
Scans, by a small margin. With a doctor able to override, 53% accepted AI reviewing images and 49% accepted AI medication decisions. Without any doctor involved, those figures were 13% and 11%.
Related reading
