Americans favor human review of healthcare AI, new Pew survey finds

Americans are willing to give AI a role in healthcare, but few want it working alone. A Pew Research Center survey published October 29 finds broad openness to human-reviewed assistance, yet 60% remain uncomfortable with their doctor using AI for diagnosis and treatment. Accepting a tool is not the same as trusting it with care.

Pew surveyed 3,488 U.S. adults from June 22–28, 2026, through its American Trends Panel. The study represents the U.S. adult population, though its findings reflect views collected four months before publication. It asked about specific healthcare tasks as well as broader expectations for patient outcomes—not whether AI actually improves medical care.

An age divide shifts since 2022

Overall expectations have changed little since Pew last asked in 2022. Today, 37% say AI used for diagnosis and treatment will improve patient outcomes, while 36% expect worse outcomes. Another 26% expect little difference. That near-even national split masks movement in opposite directions among younger and older adults.

Adults under 30 were more positive than negative in 2022; now the balance is reversed. The share expecting better outcomes fell from 44% to 31%. Adults 65 and older moved the other way: 42% now expect improvement, up from 34%. Still, 26% of that older group expect worse outcomes, and 30% anticipate little difference.

Acceptance depends on the job

Pew distinguished between AI acting alone, AI receiving a human’s final review, and no AI involvement. Translation drew the broadest acceptance: 78% were open to some AI role, including 52% with human review and 26% without it. Scheduling appointments also drew majority support, but more people favored a final human check than autonomous handling.

• Medical scans: 60% were open to AI analysis with human review, compared with 6% without it; 26% opposed any AI involvement.
• Lab results: 53% accepted AI explanations with human review, while 8% accepted AI acting alone; 31% rejected an AI role.
• Diagnosis: 47% were open to human-reviewed AI, but just 3% accepted AI making a diagnosis on its own; 41% opposed AI involvement.

Insurance decisions faced the strongest resistance of the eight tasks. Forty-six percent wanted AI excluded from claim decisions, compared with 38% open to its involvement with human review and 5% accepting independent decisions. The responses distinguish opposition to unsupervised automation from opposition to using AI at all.

Acceptance also varied by income and education. Across all eight tasks, adults with higher incomes and education levels were more receptive than those at the lowest levels. For prescription refills, for example, 86% of upper-income adults accepted some AI involvement, compared with 59% of lower-income adults.

Hospitals’ assistance model meets patient unease

Houston Methodist’s approach illustrates the distinction between assistance and authority. Michelle Stansbury, its associate chief innovation officer and vice president of IT applications, told The National News Desk that the hospital uses AI to assemble patient histories, help doctors answer portal questions and take notes by listening during appointments.

Stansbury said these tools can leave physicians more time for direct patient care, including conversation rather than typing during a visit. She emphasized that the hospital intends to keep people involved in reviewing AI outputs, rather than delegate decisions to the technology.

Patients’ expectations point to a possible tension with that promise. Forty-seven percent of respondents expected AI to worsen the patient-doctor relationship, while more than half anticipated greater vulnerability for personal health information. Doctors nevertheless retained a measure of trust: 60% had at least some confidence they would use AI in patients’ best interests.