Your doctor may already be using AI. Only 1 in 4 Americans are comfortable with that.

Americans disagree about plenty of things. Artificial intelligence in healthcare, surprisingly, may not divide them along the usual lines. When we examined a nationally weighted survey of U.S. adults, we found no statistically significant differences in attitudes toward healthcare AI by political affiliation, insurance status, or overall health after accounting for multiple comparisons.

Instead, two other factors repeatedly surfaced: education and experience with AI itself.

The findings, published in BMJ Digital Health & AI, suggest that Americans remained cautious about healthcare AI overall: only about one-quarter said they were comfortable with healthcare providers using it, and a staggering 88.4% said a healthcare provider would be better than an AI chatbot at helping them understand their care plan. Even so, the survey respondents were not neatly divided into enthusiasts and skeptics. A sizable group has yet to decide what it thinks.

That uncertainty can be just as important as outright trust or distrust. Nearly one-third of respondents were neutral about their healthcare provider using AI. Forty-four percent were neutral about physicians using AI to generate second opinions, while 40.3% chose the midpoint when asked whether they trusted AI more or less than clinicians for health information.

For us, as researchers trying to understand how patients will respond as AI becomes increasingly routine in medicine, those numbers leave a large portion of the public potentially still persuadable by experience, whether good or bad.

Familiarity may shape acceptance

One of the study’s strongest patterns involved people who had already used AI for healthcare. About one-third of respondents reported doing so. Those users consistently viewed clinical applications of AI more favorably than people without prior healthcare AI experience.

For example, 36.6% of prior users said they trusted AI “a lot” or “a great deal” to interpret laboratory or imaging results, compared with 10.4% of non-users. And 55.5% of previous users considered AI chatbots very or extremely useful, compared with 17.9% of those who had not used them for healthcare.

The study cannot determine which came first. People who are already comfortable with AI may simply be more likely to use it. Because the survey captured respondents at a single point in time, we cannot conclude that exposure itself caused attitudes to change.

Even so, the relationship raises questions as AI becomes more common in healthcare: Will familiarity make patients more comfortable with these tools? How much will early experiences shape whether they trust them? And what happens if those experiences differ across patient populations?

An education gap comes into focus

Educational attainment produced another noticeable divergence in acceptance. Among college graduates, 46.3% favored physicians using AI to generate second opinions. Among respondents without a high school diploma, only 14.6% did. College graduates were also more likely to have previously used AI for healthcare, 46.5%, compared with 22.9% among high school graduates.

With these findings comes a concern about equity: If patients with more education or greater digital literacy are the first to understand, trust, and benefit from AI-enabled healthcare, new technologies could inadvertently widen gaps they are sometimes promoted as helping to close.

We argue that equitable adoption may therefore require more than simply making AI tools available. It could require engaging communities with less exposure to digital health technologies and designing implementation strategies with them rather than for them.

A divide that largely disappeared

The final analysis also complicates one result from our earlier look at the data. While initial descriptive findings suggested political differences in attitudes toward healthcare AI, that association disappeared once the formal peer-reviewed study methodology applied survey weighting and corrected for multiple comparisons. This lack of a significant link between political affiliations and the primary outcomes was one of our more surprising findings

Insurance status also showed no significant differences, another result we found noteworthy given prior research suggesting uninsured adults may be more likely to turn to AI for health information. In our sample, uninsured respondents did not show greater acceptance of provider-deployed AI.

A public still making up its mind

The study suggests that Americans may be open to the use of AI in healthcare, but not as a replacement for human involvement in medical decision-making. Most respondents preferred speaking with a clinician before being automatically referred for additional testing on the basis of an AI finding.

The pattern extended beyond chatbots. Opposition to AI-assisted emergency triage was common across age groups, suggesting that patients may be more comfortable with AI in a supporting role than with systems making more autonomous decisions. This distinction should not be overlooked as AI becomes a more routine part of care. Patients will encounter these systems in exam rooms, patient portals, imaging results, referrals, and other parts of healthcare, often before they have formed strong opinions about them.

If anything, the survey may underestimate public skepticism. Because participants came from an online panel, people with less digital access or experience may be underrepresented. Given that greater technology exposure was associated with more favorable views of healthcare AI, the broader public could be even less comfortable than these results suggest. That leaves us with a moving target. Future studies will need to track whether familiarity changes trust, which experiences make patients more or less comfortable, and whether those patterns differ across populations.

As healthcare AI evolves, public expectations may also change with time. That makes it all the more important to ask patients what they want from these technologies, and to keep asking. Patient perspectives are still too often missing from the studies, evaluations, and implementation decisions that shape how AI enters clinical care. Measures of accuracy and performance matter, but they cannot tell us whether patients will trust a system, understand how it is being used, or feel comfortable with the role it plays in their care.

AI can only go so far in healthcare without that trust. As these tools move from research settings into exam rooms, patient portals, referrals, and clinical decisions, patients should not be treated simply as the end users of technologies designed around them. Their voices should help shape how those technologies are built, evaluated, and introduced from the outset.