AI Disagreements Could Undermine Patient Trust in Doctors

UNIVERSITY PARK, Pa. — Patients’ trust in physicians may rise or fall depending on what artificial intelligence (AI) says during a mental health visit, according to new research from Penn State and collaborators. The study shows that when an AI “second opinion” agrees with a human doctor’s recommendation, patients judge the doctor as more credible. But when the AI contradicts the doctor, patients report greater perceived uncertainty and even interpret the physician’s behavior as less diligent.

The team focused on a realistic scenario: during an online consultation, a doctor consults an AI system and shares the result with the patient. Historically, asking for a second opinion meant seeing another clinician. Now, some telehealth workflows can embed AI as an additional “expert” in the same session. The researchers asked how this changes perceptions of medical authority and doctor professionalism.

To test this, the researchers built an AI-powered chatbot that roleplayed as a human doctor named Dr. Alex. Participants first received a brief cognitive behavioral therapy (CBT)-style consultation about everyday stressors. Dr. Alex concluded that CBT was appropriate, and then patients were offered an optional AI second opinion.

For the second opinion, the researchers used a chatbot called CareBot. In different conditions, CareBot either agreed or disagreed with Dr. Alex’s recommendation. The assistant delivered the exchange in a way designed to mirror conversational clinical guidance, enabling consistent presentation of the same medical context across participants.

The experiment recruited 135 adults in the United States and then measured post-session evaluations of doctor credibility and perceived medical certainty. The results revealed an asymmetry: agreement between Dr. Alex and the AI increased patients’ confidence. Disagreement amplified judgments of uncertainty and “doctor laziness.”

Importantly, these effects depended on anthropomorphism—how human-like the doctor appeared through the interface. The study found that positive and negative consequences of AI agreement or disagreement emerged only when participants perceived the doctor as sufficiently human. In other words, the same technical contradiction can be socially interpreted very differently depending on perceived identity cues.

The findings carry direct implications for telehealth platforms and app-based medicine, where clinicians’ real identities may be obscured and AI co-present. More than half of participants in the study treated the AI-simulated doctor as human-like, suggesting that conversational behavior alone can create strong authority.

The negative impact of AI disagreement was strongest among individuals who believe machines are more objective and precise than humans. For those patients, AI contradictions can plant doubts rapidly, shifting trust away from the physician.

Finally, the research proposes communication strategies for clinical settings: doctors can acknowledge AI differences implicitly, invite joint exploration of the conflicting points, and explain plausible reasons for disagreement, such as data limitations affecting whether results generalize to different patient groups.