West Health–Gallup finds patient comfort with medical AI drops sharply without clinician review

Patients are far more comfortable letting AI make diagnosis or treatment decisions when a clinician can intervene. Findings released November 10 by the West Health–Gallup Center on Healthcare in America put comfort at 56% when providers could change or override those decisions, compared with just 9% when AI acted without provider review.

The patient findings come from an August 12–24, 2026, Gallup Panel survey of 5,330 U.S. adults. Gallup recruits panelists through probability-based random sampling. The margin of sampling error for the full sample was plus or minus 1.7 percentage points, at a 95% confidence level. Results on comfort with diagnosis and treatment decisions cover respondents who had seen a healthcare provider in the past year.

Three roles for AI, three levels of human involvement

Rather than ask only whether patients accepted medical AI, the survey separated three ways it could participate in care. Each assigned a different degree of authority to the technology:

• AI assists: The system recommends an action, but the provider makes the final decision.
• AI decides with oversight: The provider can change or override the system’s decision.
• AI decides independently: The system makes the decision without provider review.

The same boundary appeared in imaging and medication decisions. For diagnostic images, 53% were comfortable with AI decisions a provider could override, versus 13% without review. For medication decisions, the figures were 49% and 11%. When AI only assisted and the provider decided, comfort was 52% for imaging and 45% for medications.

Even patients who trusted AI-generated health information preferred clinician involvement. They made up 23% of patients. Among them, 65% to 71% were comfortable with AI across the clinical uses surveyed when a provider could review its work. Without review, comfort ranged from 18% to 20%. Separately, 39% of patients said they had used AI in the past year to seek physical or mental health information or advice.

Providers underestimate comfort, while training trails use

The companion provider survey found a mismatch in expectations. Only 36% of providers believed most of their patients would be comfortable with AI making diagnosis or treatment decisions they could override. That was 20 percentage points below the share of patients expressing comfort. The comparison concerns providers’ expectations about patients, not their own willingness to use the technology.

Unlike the probability-based patient panel, the provider study used a nonprobability, opt-in panel of 1,007 respondents, surveyed August 7–23, 2026. Its methodology describes web surveys conducted in English, excluding people without internet access, and reports a credibility interval of plus or minus 3.3 percentage points at the 95% confidence level.

Providers most often said they used AI to gather information about medications (78%) or symptoms and conditions (76%). Many also used it for patient-specific work: 53% reviewed information such as test results or imaging, and 50% wrote clinical notes. Nearly half used AI for patient instructions or after-visit summaries, to retrieve or summarize medical records, or to record and transcribe appointments.

Training varied substantially by workplace. Thirty percent of providers in hospitals or health-system practices reported formal training, compared with 16% in provider groups and 11% in independent practices. Even in the highest-training setting, fewer than one-third reported receiving it.

A route back to a clinician

For patients uneasy about AI acting alone, the ability to request human review offered a specific reassurance. Among those uncomfortable with diagnosis or treatment decisions without oversight, 48% said that option would increase their comfort. For those uncomfortable with unreviewed imaging interpretation, 47% said the same, according to CIDRAP’s account of the findings.

Other proposed reassurances drew less support from those uncomfortable patients. Fewer than one-third said strong privacy protections, evidence of safety and effectiveness, faster decisions or lower healthcare costs would make them more comfortable. These answers measure what patients say would reassure them, rather than testing whether any safeguard improves clinical care.