Providers Underestimate Patient Comfort With AI in Care

Editor’s Note: This research was conducted in partnership with West Health through the West Health-Gallup Center on Healthcare in America, a joint initiative to report the voices and experiences of Americans within the healthcare system.

WASHINGTON, D.C. — More than half of U.S. adults (56%) who have seen a primary care provider in the past year are comfortable with artificial intelligence making diagnosis or treatment decisions, as long as their healthcare provider can change or override those decisions. By contrast, just 9% are comfortable when AI makes these decisions without provider review. The same pattern holds with respect to the use of AI in imaging reviews and for medication decisions: Patients are much more comfortable with these applications of AI when a provider remains involved.

These findings are based on an Aug. 12-24, 2026, probability-based Gallup Panel web survey of 5,330 U.S. adults who have seen their primary care provider in the past year. These adults are referred to as “patients” throughout this article.

Patient Comfort With AI Varies by Level of Provider Involvement

Patients were asked about their comfort with AI use in three aspects of clinical care: diagnosis and treatment, imaging review, and medication decisions. For each application, patients were asked about their comfort with three levels of reliance on AI in patient care decisions: having providers use input from AI when making their own patient-care decisions, having AI make decisions but giving providers the option to override those decisions if they disagree, and having AI make decisions without provider review.

Patients are most comfortable with AI when a provider remains involved, and comfort drops sharply without provider review.

For diagnosis and treatment, 56% of patients say they are somewhat or very comfortable with AI making decisions that their healthcare provider could change or override, compared with 38% who say they are comfortable when AI assists and the provider makes the decision. Just 9% are comfortable with AI making diagnosis or treatment decisions without provider review.

Across all three clinical uses, patient comfort is highest when AI makes the decision but the provider can change or override it. However, the difference in comfort between this arrangement and one in which AI assists and the provider decides is much smaller for imaging review (53% vs. 52%) and medication decisions (49% vs. 45%) than for diagnosis and treatment (56% vs. 38%). When AI makes decisions without provider review, comfort falls to 13% for imaging review and 11% for medication decisions.

Even among the subset of patients who are generally receptive to AI for healthcare information, comfort is substantially higher when a provider remains involved. Among the 23% of patients who say they somewhat trust (21%) or strongly trust (2%) the accuracy of AI-generated health information, 65% to 71% are comfortable across the three clinical uses when AI assists the provider or makes decisions the provider can override, compared with 18% to 20% when AI makes decisions without provider review.

A similar pattern holds among the 39% of patients who have personally used AI to gather physical or mental health information or advice in the past year. Across the three clinical uses, 51% to 66% are comfortable with either form of provider involvement, compared with 12% to 16% when AI makes decisions without provider review.

Provider Review Matters More to Patients’ Comfort Than Cost or Speed

Patients who are uncomfortable with AI making diagnosis and treatment decisions or reviewing medical images without provider review were asked whether five potential factors would make them more comfortable with their provider relying exclusively on AI, would make them less comfortable or would make no difference.

No single factor is selected by a majority of patients as making them more comfortable with AI making autonomous decisions about their healthcare, but the most helpful is the ability to request provider review, selected by just under half in both scenarios.

Of the other factors tested — having strong privacy protections, evidence of AI’s safety and effectiveness, lower healthcare costs and faster decisions — none would make more than a third of these patients more comfortable. 

What Primary Care Providers Think and How They Use AI

A separate non-probability opt-in survey of 1,007 licensed, practicing U.S. primary care physicians with MD or DO degrees (referred to as “providers” in this article) was also conducted in August 2026 to complement the patient findings.

Provider Estimates Sometimes Differ From Patients' Own Responses

Providers were asked whether they believe most of their patients would be comfortable with the same clinical AI applications and levels of provider involvement examined in the patient survey.

Providers generally expect patients to be least comfortable with AI making clinical decisions without provider review, consistent with patients’ own responses.

Across all three clinical applications, providers underestimate patient comfort when AI makes decisions that a provider can change or override. The gap is largest for diagnosis and treatment, where 56% of patients say they would be comfortable with this arrangement, 20 percentage points higher than the share of providers who say most of their patients would be. Differences are smaller for imaging review (53% of patients vs. 42% of providers) and medication decisions (49% vs. 42%). When AI assists and the provider makes the decision, provider estimates are within two points of patients’ responses for diagnosis and treatment and for medication decisions, but providers still underestimate patient comfort with imaging review by 11 points.

Most Providers Report Using AI, but Few Have Formal Training

Nearly nine in 10 providers (89%) report personally using an AI tool or chatbot for at least one of seven surveyed work tasks in the past year, including about two-thirds (66%) who report using AI for at least one task weekly or more often.

Looking at individual tasks, the most common uses in the past year are gathering information about medications (78%) and gathering information about symptoms or health conditions (76%).

AI use extends beyond information gathering. Over half of providers (53%) report using AI to review patient-specific information such as test results, scans or imaging, while 50% have used it to write clinical notes. Nearly half have used AI to write patient instructions or after-visit summaries (49%), summarize or find information in medical records (47%), or record and transcribe patient appointments (45%).

Despite widespread work-related use, only 20% of providers report receiving any formal training in using AI in healthcare settings. Reported training varies by practice organization, from 30% among providers in hospital or health system practices to 16% in physician groups and 11% in independent practices.

Implications

AI's role in healthcare is evolving rapidly, with patients turning to AI for health information and healthcare organizations exploring and implementing AI across a growing range of applications. Previous West Health-Gallup research found that Americans who use AI for health information generally do so to supplement traditional healthcare, although some also report skipping provider visits after receiving AI-generated information or advice. For the first time, the West Health-Gallup Center on Healthcare in America examined patient and provider perspectives on AI in healthcare through complementary surveys, offering new insight into how patients' opinions compare with providers' perceptions of those opinions.

This comparison reveals a potential disconnect between patients' reported comfort and providers' perceptions of that comfort. Providers may underestimate patients' comfort in some situations when AI makes clinical decisions that providers can override, particularly for diagnosis and treatment. These differences suggest that patients may be more receptive to certain forms of AI than providers realize, reinforcing the value of asking patients about their preferences directly. Misjudging patients’ comfort could mean missed opportunities for healthcare organizations to explore potentially beneficial AI applications that align with patient preferences.

At the same time, AI is already widely used by providers for a range of work tasks, even as relatively few report receiving formal training in healthcare settings. With AI adoption already underway, an important question is what training, guidance or other support would best help providers use AI effectively and responsibly, particularly when it informs clinical decisions, and how these needs may differ across tasks and practice settings.

Patients are considerably more comfortable with AI making decisions when providers can intervene than when there is no provider review. However, the form of provider involvement also matters. For diagnosis and treatment, patients are even more comfortable when AI makes decisions that providers can override than when AI assists providers who make the decisions. As AI capabilities and applications in healthcare continue to evolve, tracking how patient comfort and provider perceptions change over time, across different clinical uses and levels of provider involvement, will be important for understanding how healthcare organizations can best incorporate AI in ways that reflect patient preferences.

Explore more of the data and insights at westhealth.gallup.com.

Stay up to date with the latest insights by following @Gallup on X and on Instagram and @West Health on X and LinkedIn.