What Patients Think of AI-Assisted Health Care Communication

Today’s patients increasingly find themselves on the receiving end of AI-integrated health care. From AI-drafted chart messages to chatbots handling intake and triage, patients are largely aware that not every digital health care interaction has a human on the other side, and opinions on this practice vary widely. How are patients navigating increasingly AI-integrated health care services, and what are their feelings about this change?

Patients may value AI’s efficiency while still having concerns over privacy, reliability, and loss of human judgment. Samantha Hovley sees both sides: she’s a health care IT professional who’s also experienced the patient’s perspective after being diagnosed with a rare disease in 2017. As part of her ongoing care, she often gets AI-drafted messages in her care portal. “I don’t mind AI-drafted messages from my care team,” Hovley shares. “I suspect that puts me in the minority.”

She now runs a patient advocacy TikTok account called Patient Strategy, where she breaks down how the US health care system works. In one video, she details recent changes to Epic, a widely used electronic health care system. Some of these changes incorporate AI into patient messaging and care workflows more heavily. Hovley says viewers’ comments on the video carried a common theme: “The comment section turned into a raw read on why patients object. Most of it wasn’t about accuracy. It was about whether their doctor actually read the message.”

Contextual Factors Influencing Patient Preferences for AI-Assisted Communication

AI acceptance in the health care setting varies broadly between individuals and groups. One survey reported that factors such as patient age, race and ethnicity, education level, gender, and household income all impacted receptiveness, rates of use, and trust in AI tools for health care.

Patients’ trust in health AI varies most strongly based on the context in which it’s used. AI trust scored highest in clinical and administrative areas, with 50% of respondents agreeing that they trust AI to perform mundane tasks such as chart summaries and clinical notes, so long as there is human oversight. That number drops by nearly half for more complex situations such as insurance applications, with only ~28% of survey participants expressing comfort with AI-reviewed insurance claims, even if human oversight was included. Another study on patient perceptions of AI scribes echoes the conclusion that acceptance depends heavily on when and how AI is used, as well as transparency, privacy, consent, and clinician oversight.

‎

Individual opinions reflect this divided stance. Respondents shared their thoughts on an informal social media post on the topic:

“The key to patient safety is the sender and receiver must be able to discern this information [is AI-generated]. This key feature is with the human.”

“AI-generated responses...never answer my actual, nuanced questions. A waste of everyone’s time and energy.”

“I would think that AI-generated responses would be so generalized that they wouldn’t be helpful to each patient and their unique situation.”

Emily Hartstone—a patient advocate who lives with an autoimmune disease herself—summarizes the tension. “My opinion isn’t simple. The speed [of AI chart messaging] is real and I’ve been glad of it more than once. But I can always tell, and that turns out to matter. Once you can spot which messages a machine drafted, a question goes over all of them, including the ones a person wrote at eleven at night thinking about your case.”

Further Insights on AI Health Communication

Patient preferences on health AI vary according to the context and stakes of the interaction. Routine communication may be an area where patients are particularly receptive to AI assistance, while more complex, sensitive, or consequential situations create different expectations for clinician involvement and communication.

Kellie Owens, PhD, is an assistant professor of medical ethics at the NYU Grossman School of Medicine. She coauthored a report on patient perceptions of AI-drafted chart messages. Owens says of her team’s findings, “Patients in our study prioritized quick answers to their questions over relational depth. If clinicians and health systems can use AI to aid with portal messaging in ways that improve response times while still maintaining human review, that seems to align with patient preferences.”

While some respondents found that AI messages were too long or formulaic to be helpful, others described the supportive phrasing as reassuring and thorough. Patients did not necessarily view every portal interaction as requiring a highly personalized or empathetic response, though in some situations, a clearly human answer was preferred. Owens elaborates that “human review was integral to maintaining patient trust...we need to design our online messaging systems in ways that ensure meaningful human review and editing.”

Human Oversight, Accountability, and Patient Trust

The conditions that make patients comfortable with health care AI depend on both personal factors as well as how and when AI is used. Clinician review was a key component of patient acceptance. Patients may interpret clinician review as evidence that someone knows their history, understands their question, and is accountable for the response. Hovley’s TikTok audience, as well as Owens’s study, make it clear that there is a marked difference in acceptance of cases where AI simply helps a clinician respond more efficiently vs cases where the response creates the impression that no human has actually read the message.

As health systems develop policies around AI-generated patient messages, transparency and disclosure must remain top of mind to maintain patient trust. Owens’s findings supported this: “Our study population consistently wanted AI use in portal messaging to be disclosed.”

Disclosure policies should consider whether to focus solely on informing patients that AI was used or whether to make clear how and when human oversight is included. Depending on context, clinicians may also be able to communicate these details directly to their patients. Implementation should account for clinician accountability, the stakes and context of the communication in question, appropriate tone and level of detail, and patient expectations for human involvement.

Maintaining Human-Centered Communication in AI-Integrated Care

For health systems, building trust in AI-assisted communication may depend less on making AI sound human and more on making human oversight visible and meaningful. Patients may accept AI when it makes health care more responsive and efficient, but acceptance fades when AI creates uncertainty about who is responsible for the communication or whether the clinician meaningfully engaged with the message. As health care continues to move toward the future and both health systems and patients adapt to changing technologies, open disclosure and the human touch remain key to maintaining trust.