Patients would rather wait for a clinician’s review than get an instant AI answer to medical questions

Given a choice between an artificial intelligence (AI) answer to a medical portal message within five minutes or the same AI answer reviewed and edited by a doctor or nurse within 48 hours, 63.1% of U.S. adults chose to wait, according to a survey study published Oct. 14 in JAMA Network Open.

The preference flipped for routine requests. For logistical messages such as appointment scheduling and prescription refills, 57.4% of respondents wanted the faster AI response. For general messages, including nonmedical questions and administrative requests, the split was close to even, with 50.6% choosing speed.

Researchers from Columbia University School of Nursing and Vanderbilt University Medical Center surveyed 1,185 adults recruited through Qualtrics online survey panels between June 17 and July 26, 2024. The sample was balanced on gender, education and generation, and researchers targeted 50% of participants identifying as members of racial or ethnic minority groups.

“When people have a medical question, they want to hear from their doctor, not AI, even if AI may give an immediate answer,” lead author Natalie C. Benda, Ph.D., assistant professor at Columbia Nursing, said in a news release.

The slower option still used AI

The survey did not pit AI against a human-written reply. Participants chose between a fast answer from the AI within five minutes and “waiting until a health professional (doctor or nurse) reviews and edits the AI’s answer (up to 48 hours).”

A 2025 Duke University Health System survey of 1,455 members of its patient advisory committee, also published in JAMA Network Open, found respondents slightly preferred AI-drafted portal replies over clinician-written ones when they were not told who wrote them. Satisfaction dropped by 0.13 points on a 5-point scale when respondents were told AI wrote the reply rather than a human. More than 75% were satisfied regardless of author or disclosure.

The Duke team found no sign that the seriousness of the message, which ranged from a routine refill request to a possible cancer on imaging, changed how patients reacted to being told AI was involved. Its authors theorized that patients had not yet developed preferences about which situations call for AI.

The Columbia survey, fielded about six months after Duke’s closed, asked patients to trade speed against review instead, and their answers split by message type.

Who was more open to AI

Black participants, who made up 39.3% of the sample, were 1.4 times as likely as White participants to accept a fast AI answer to a medical question, and more likely to accept AI for general and logistical messages as well. Women were about 24% less likely than men to accept AI for medical questions.

Participants with higher health literacy were more open to AI for general and medical questions. Participants who reported more discrimination in medical settings were modestly more likely to accept AI answers to medical questions. Generation Z participants were less likely than Baby Boomers to accept AI for logistical messages.

A human therapist, even three months out

On mental health support, 63.9% of participants said they were likely to use a licensed human therapist available in three months. Half, 50.2%, said the same of an AI chatbot monitored by a licensed therapist and available within a month, and 44.3% of an AI chatbot alone, available immediately.

What review costs a practice

Patient-authored portal messages rose from 0.99 per patient per year in 2020 to 2.5 in 2025, according to an analysis of Epic Cosmos data published in JAMA in June.

At Mass General Brigham, 77.5% of portal message threads sent to primary care clinicians drew a response from the care team, but 32.0% drew one from the clinician the patient had written to, according to an analysis of more than 3.6 million messages published in August.

In an invited commentary accompanying the Columbia study, Felix Busch, M.D., of TUM University Hospital in Munich, and Dennis M. Hedderich wrote that human oversight consumes clinicians’ time and delays information and support patients value.

“Safety should determine the minimum level of oversight, while patient preference may guide additional review,” they wrote. They added that health systems should prioritize equal access to human care and escalation rather than use demographic characteristics to steer patients toward automated services.

Marc Succi, M.D., associate professor of radiology at Harvard Medical School and executive director of the Mass General Brigham MESH Incubator, told Medical Economics in April that he would be comfortable using AI for low-risk tasks such as documentation, summarization and billing. He put responding to patient messages on his list of higher-risk uses.

“That is where you need to stop and look at the level of performance very critically and in multiple ways, not just trust what the vendors say,” Succi said.

Asked how physicians keep that review from becoming a formality, Succi said, “The human cannot just rubber-stamp the output, but the real issue is liability. How much risk is that physician willing to take?”

The Columbia survey grouped prescription refill requests with appointment scheduling as logistical messages, the category where most respondents chose speed. Succi put renewing medications on his higher-risk list. In Utah, Doctronic’s AI platform began renewing prescriptions for patients with chronic diseases without an office visit in January under a state-sanctioned pilot. The Utah Medical Licensing Board demanded an immediate shutdown in April, and state officials had kept the program running as of late May.

Limits of the study

Because the survey tied response time to who reviewed the answer, it cannot separate a preference for human review from a reaction to the wait, the authors wrote. The scenarios were hypothetical, survey wording may have framed answers, the sample was not a probability sample, answer options were not randomized and only U.S. adults were surveyed. How people interact with AI has also changed since the data were collected in mid-2024.

The researchers called for further study of offering patients a choice between faster responses and human review.

One author, Xuhai Xu, Ph.D., of Columbia’s Department of Biomedical Informatics, reported receiving personal fees from Google during the conduct of the study.