What do patients think of AI in patient portal messages?

When it comes to AI in patient portal messaging, patients have their limits. Aa new JAMA Network Open report shows they strongly prefer the technology be deployed to manage low-level requests, not clinical needs.

The report, based on a survey of 1,185 patients, indicates that AI might not be a cure-all for providers' growing patient portal inboxes.

Between 2020 and 2025, patient portal message volume increased by 153%, according to a separate JAMA report. All said, patient portal message volumes can yield up to 2.6 hours of after-hours work for providers.

Generative AI is a promising solution to patient portal workloads. Many hospitals and health systems have implemented the technology to help draft provider replies, with best practices stating that healthcare professionals should review every AI-generated message before it is sent to patients.

But according to this latest JAMA report, patients would prefer to hear from their clinician, not AI, when discussing their clinical needs.

The researchers surveyed patients about which types of patient portal messages are best suited for AI responses, posing their questions as a choice between AI's speed and their clinician's human-led expertise.

More than half (57.4%) said they preferred a faster, AI-generated response to logistical questions, like appointment scheduling or prescription refills. Another 50.6% preferred quicker AI messages for general questions, like nonmedical inquiries or administrative requests.

But when it comes to medical questions, such as presenting a new symptom or a treatment question, patients wanted to hear from clinicians directly, with 63.1% saying they'd be willing to wait for a healthcare professional.

These trends persisted in mental health settings.

More than half of patients (63.9%) wanted to hear directly from a therapist when seeking mental health support, while 50.2% were comfortable with therapist-reviewed AI messages and 44.3% were comfortable chatting with AI alone.

According to the researchers, context and time constraints predict a patient's acceptance of AI in patient portal messaging. Respondents were receptive to AI  for low-stakes but time-sensitive tasks, like booking an appointment or getting a prescription refill.

However, when clinical content came into the mix, patients increasingly wanted their providers to at least review an AI-generated patient portal response or to receive a response directly from the provider.

"Our findings warrant further investigation in offering patients options for tradeoffs between timeliness and human review of generative AI content," they concluded.

These findings might influence how healthcare organizations deploy AI in the patient portal. For example, they might consider segmenting what types of messages -- low-level tasks versus complex clinical requests -- receive AI responses.

It'd also be useful to understand the true impact AI has on providers' workloads. Some studies have shown that AI-drafted patient portal messages can make providers faster to respond to messages.

However, other studies have shown AI doesn't move the needle much on cognitive burden. In fact, AI can increase cognitive burden because it can often misalign with provider responses and require rewrites.

Sara Heath is an executive editor at Xtelligent Healthcare Media, where she covers patient engagement, healthcare policy and health IT.