Patients Are Open to AI. They Want a Human Accountable for It.
New Pew data shows that Americans will accept AI in care, on one condition: a person stays responsible for the decision.
The public has not rejected healthcare AI. It has set terms. In Pew Research Center's new survey of 3,488 U.S. adults, 78% are open to AI translating for patients, and most of them want a person reviewing the result. Only 5% would let AI decide insurance claims with no human review.
For anyone deploying AI in a health system, that is less a sentiment finding than a requirement. Patients are describing the oversight they expect.
Openness has a condition attached
Americans separate AI that assists from AI that decides. Human review is the most common answer for every task except insurance claims, where outright rejection leads.
Healthcare task | Open to AI alone | Open with human final review | Not open |
Translating for a patient | 26% | 52% | 12% |
Scheduling an appointment | 29% | 44% | 18% |
Ordering prescription refills | 17% | 53% | 21% |
Taking appointment notes | 13% | 56% | 23% |
Analyzing medical scans | 6% | 60% | 26% |
Explaining lab results | 8% | 53% | 31% |
Making a diagnosis | 3% | 47% | 41% |
Deciding an insurance claim | 5% | 38% | 46% |
% of U.S. adults. Other responses not shown. Source: Pew Research Center survey, June 22-28, 2026.
Openness drops as the stakes rise. Scheduling and translation draw the most support for AI working alone, while diagnoses and claims draw the least.
Personal comfort is lower still. Six in 10 adults say they would be uncomfortable with AI in their own diagnosis and treatment; 30% would be comfortable. On patient outcomes, the public is evenly split: 37% expect better results, 36% expect worse, and 26% expect little difference. Pew says that split has barely moved since 2022.
Trust is uneven, and the gaps are widening
The national average hides sharp divides. Upper-income adults are more than twice as likely as lower-income adults to expect better outcomes, 55% versus 25%. The pattern holds across all eight tasks: 86% of upper-income adults are open to AI ordering refills, compared with 59% of lower-income adults.
Age cuts the other way from what many would assume. Among adults under 30, 45% now expect worse outcomes, up from 29% in 2022, while 31% expect better. Adults 65 and older are the more optimistic group, with 42% expecting better outcomes, up 8 points.
Concern is also rising among Hispanic and Black adults, whose shares expecting worse outcomes grew 11 and 12 points since 2022. For teams deploying AI across diverse populations, that trend deserves attention.
The specific worries point to governance more than to clinical performance:
Impact of AI on... | Worse | Better |
Security of health information | 51% | 8% |
Patient-doctor relationship | 47% | 10% |
Healthcare costs | 26% | 19% |
Access to care | 20% | 31% |
Between 22% and 32% were unsure on these measures, so many opinions are still forming. Americans also doubt the institutions meant to keep AI in check. At least half express low or no confidence in insurers, the federal government, AI companies and state governments to ensure AI serves patients. For doctors, 26% express a great deal or quite a bit of confidence, and 34% express some.
What this means for healthcare AI leaders
We read these numbers as a governance brief written by patients. Four priorities stand out.
Match oversight to stakes. The public already sorts AI by risk. Your inventory should too, with the tightest controls on tools that shape diagnoses, coverage and treatment.
Make human review real. A final review only counts if the reviewer has the time, context and authority to overrule the model. Define who reviews what, and measure how often they disagree.
Monitor performance by population. Expectations differ sharply by income, age and race, so an average result can hide who is being served poorly. Track outcomes across subgroups continuously, not only at launch.
Treat data security and the care relationship as adoption issues. These are the two areas where the public most expects harm. Tell patients how AI is used, what data it touches and who is accountable.
The bottom line
Patients are not asking healthcare to slow down on AI. They are asking who is responsible when it is used on them. With public confidence this thin, health systems will have to earn it through oversight they can demonstrate.
That is the work we focus on at ALIGNMT AI: helping healthcare organizations govern and monitor AI so that accountability is visible, not assumed.
About the data: Pew surveyed 3,488 U.S. adults from June 22 to 28, 2026, through its American Trends Panel. The margin of error for the full sample is plus or minus 1.8 percentage points.
Source: Americans Are Open to AI in Healthcare, but Want Humans Making Final Calls, Pew Research Center, October 29, 2026.
Comments