What Patients Actually Want From the Message
AI-drafted patient-portal messaging is judged not only by the clinician who releases it but by the patient who must use it. The key question here is what patients are trying to accomplish in the portal, and how that purpose changes the value of speed, brevity, detail, empathy, and even the choice of communication channel. The available qualitative evidence gives readers a way to evaluate a generated reply by fit to the patient’s situation rather than by whether it sounds uniformly human.
This matters because “patient experience” is often treated as if it had one direction. A warmer, longer message is assumed to be more caring; a faster message is assumed to be more convenient; disclosure of AI is assumed either to build trust or to destroy it. The interviews show a more conditional picture. Patients may want an efficient transaction in one setting and a careful conversation in another. They may appreciate courteous language in one message and find the same language artificial in the next. Those differences are not noise around a single preference. They are evidence that communication quality depends on what the message is for.
The portal is often a place to solve a problem, not recreate a visit
The qualitative study interviewed 40 adult patients from a large academic health system. Most participants described portal messaging as useful for quick information, problem resolution, refills, and simple exchanges. They generally did not treat it as a setting for deep interpersonal connection. Several participants made the point in practical terms: they were not opening the portal to be comforted in the way they might expect during an office visit; they wanted an answer and a way to take care of the issue.
That finding changes how AI assistance should be evaluated. If a patient uses the portal to learn whether a request was received or what to do about a routine matter, a draft that helps a clinician answer clearly and promptly may improve the experience. The value is not that the machine has reproduced a relationship. It is that the communication channel has done its limited job with less waiting. Patients who were uneasy about AI in other parts of medicine could still find it acceptable in online messaging because the portal already felt more transactional than an examination room.
This distinction between a communication channel and the broader care relationship is important. A patient can accept automation for a narrow exchange while rejecting it for diagnosis, treatment decisions, or an in-person encounter. Acceptance in the portal does not transfer automatically to every other use of AI. Nor does a transactional message become irrelevant to care simply because it is not relational. A refill, symptom report, or explanation of a result can alter what the patient does next. The channel sets expectations for the form of the exchange; it does not remove the clinical consequences of its content.
“Transactional” does not mean unimportant or consequence-free. It means that patients often have a specific practical objective and a limited expectation of relational depth in that medium. A response that spends many sentences performing warmth without answering the question may fail the transaction. A response that is short but omits the information needed for a safe decision fails it more seriously.