NEW YORK CITY — A study published in JAMA Network Open by Kellie Owens, PhD, and her team at New York University Grossman School of Medicine examined patient perceptions of AI-drafted messages in healthcare portal communication. The study involved 40 adult participants and found that patients expressed comfort with AI-drafted messages but emphasized the importance of clinician review and transparency regarding AI use.

Participants engaged in 45 to 60-minute videoconference interviews. Investigators utilized vignette-based prompts to explore patient perceptions of messaging in clinical situations such as medication refill requests, side effects, and potentially abnormal test results. Participants evaluated messages associated with specific scenarios after answering open-ended questions, observing examples of patient messages paired with draft responses that varied in tone, length, and implied authorship.

Participants viewed portal messaging as effective for routine clinical needs, with most framing it as "fundamentally transactional." They considered it suitable for obtaining routine information and resolving situations quickly. However, participants did not view portal messaging as appropriate for situations requiring deep interpersonal connection.

Comfort with AI-drafted messages did not influence preferences about message tone or expressions of empathy. Perceptions of message appropriateness depended on the purpose and stakes of the communication. The study participants broadly favored disclosure of AI use, and all patients expressed a desire for transparency about AI use across healthcare contexts.

"Our study suggests that patients are generally comfortable with the use of generative AI to draft in-basket responses, particularly when framed as a workflow support that could improve response speed and reduce clinician burden," Owens said. "Patients emphasized that the primary value of portal communication is timely resolution of health-related questions rather than personalization or relational depth."

Prior surveys indicated that participants generally preferred AI-drafted responses over clinician-written responses and rated them more satisfactorily across various scenarios. However, satisfaction with AI-drafted responses was lower when participants were explicitly informed that a response was generated by AI. "This paradox -- preference for AI-drafted messages alongside reduced ratings when AI involvement is disclosed -- suggests that patient evaluations are not only shaped by the content of messages but by beliefs about what AI use means in the context of clinical relationships and delivery of healthcare," Owens noted.

Patients perceived AI technology as a reasonable response to clinician workload. They emphasized that AI drafting was acceptable only if messages were reviewed and approved by a clinician before being sent, extending even to lower-stakes tasks. Preferences for message tone and content were dependent on the clinical stakes, with some patients preferring more detailed or emotionally supportive language in higher-stakes scenarios like the interpretation of potentially concerning test results.

Teva Brender, MD, affiliated with the University of California San Francisco, commented on the findings. "As one of the first qualitative investigations on the subject, the study by Owens et al. contributes essential nuance to the growing literature on AI-drafted portal messaging," Brender said. "For now, while respondents in this study were comfortable with AI's use, disclosure and oversight remain essential."

Why It Matters

This study provides insight into patient expectations regarding artificial intelligence in healthcare communication. The findings indicate that while patients are open to AI's use for efficiency in routine health inquiries, they maintain a desire for human oversight and transparency, particularly in more sensitive medical contexts. These results could inform how healthcare systems implement AI-assisted messaging to ensure patient trust and satisfaction. The study included 75% women, 32.5% white participants, 35% Black or African-American participants, and one-third of participants were aged 65 or older.