What the study found
Patients viewed patient portal messaging as mainly transactional, with a focus on timely problem-solving rather than relationship-building. They were generally comfortable with AI-drafted replies, but that comfort depended on clinician oversight and accountability.
Why the authors say this matters
The authors conclude that AI use in portal messaging may support more efficient communication if clinicians remain responsible for the messages. They also say implementation should include clinician oversight, context-sensitive communication standards, standardized disclosure practices, and monitoring of quality, equity, patient trust, and safety.
What the researchers tested
The study used a qualitative design with vignette-based prompts modeled on scenarios from a prior survey about AI-drafted online messaging. Researchers interviewed 40 adult patients from a large academic health system by videoconference between April and August 2025, using draft messages that varied in tone, length, and implied authorship.
What worked and what didn't
Patients expressed high comfort with AI-drafted messages when a clinician reviewed them. Preferences for tone and empathy varied by person and situation, and were shaped more by whether the message fit the purpose and seriousness of the communication than by whether it seemed AI-like or human-like. Participants broadly supported AI disclosure, but they differed on when and how disclosure should be presented.
What to keep in mind
This was a qualitative study of 40 patients from one large academic health system, so the findings describe this sample rather than a broader population. The abstract does not describe long-term outcomes, and it notes that downstream effects on quality, equity, trust, and safety should be monitored.
Key points
- Patients described portal messages as transactional and focused on timely problem resolution.
- Comfort with AI-drafted replies was high when clinicians reviewed the messages.
- Message tone, length, and detail mattered more than whether a message seemed AI-like or human-like.
- Participants broadly wanted disclosure that AI was involved, but preferred different timing and formats.
- The authors recommend clinician oversight and monitoring of effects on quality, equity, trust, and safety.
Disclosure
- Research title:
- Patients supported AI-drafted portal messages with clinician oversight
- Authors:
- Kellie Owens, Athmeya Jayaram, Anand Chowdhury, Kathryn I. Pollak, Sam Klotman, Zachary Griffen, Miles Danielski, Ben Goldstein, Jennifer Maddocks, Matthew Roman, Eric G. Poon, A Bedoya, Joanna Cavalier
- Institutions:
- Duke University, Duke University, Duke University, Duke University, Duke University, Duke University Health System, Duke University Health System, Duke University Health System, Duke University Health System, Hastings Center, Hastings Center, New York University, Wesleyan University
- Publication date:
- 2026-07-07
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with scholarly research without reading academic papers — one filtered digest, every Friday.