Trust Is a Workflow, Not a Tone

Trust in AI-drafted patient-portal messaging depends on more than whether a reply sounds warm or whether a patient is told that AI was involved. The practical question is whether disclosure accurately describes the relationship among the model, the clinician, and the patient—and whether the organization actually enforces that relationship. Readers should be able to judge a disclosure practice by what it makes visible, what it leaves unresolved, and whether the surrounding workflow turns clinician accountability into a real action rather than a reassuring label.

This distinction matters because trust is often discussed as a reaction to wording. A system may try to make a message sound human, add a statement about improving the care experience, or conceal the use of AI to avoid concern. The patient evidence points in a different direction. People wanted transparency, but they also wanted the message to fit the situation and a clinician to remain responsible. A clear disclosure can support honest expectations. It cannot compensate for an incomplete reply, a missed escalation, or a review step that exists only on paper.

Disclosure should make the accountability arrangement legible

In interviews, patients broadly wanted to know when AI was used in health-care communication. Within the 40-person qualitative sample, transparency was generally associated with honesty and trust rather than automatically with skepticism. Some participants wanted to be told every time. Others preferred advance notice during the care process plus a reminder attached to the message. Many did not find one proposed disclosure statement meaningfully better than another. The broad preference was for transparency; the operational details remained unsettled.

That pattern gives disclosure a more precise job. It should tell the patient that AI helped produce the message and identify what a clinician did with that draft—reviewed it, edited it, approved it, or some combination that the institution can defend. It should not imply that AI authorship is the only relevant fact. The patient is trying to understand who stands behind the content and what kind of judgment occurred before the message arrived. A plain statement about the actual process is more informative than a vague assurance that technology was used “to improve the care experience.”

There are two different ways a disclosure can mislead. It can obscure authorship, leaving the patient to assume that a clinician personally composed every word. Or it can name AI use while implying that the presence of a clinician guarantees correctness. The first hides a material fact about how the message was made. The second turns a review expectation into a promise the workflow may not be able to keep. Honest disclosure must avoid both errors: it should describe the process without overstating what the process proves.

Plain language matters because marketing-like wording can blur responsibility. “AI assisted with this message; a clinician reviewed it” describes an arrangement the patient can evaluate. “This message was generated to improve your care experience” describes an intention while leaving authorship, review, and accountability uncertain. The difference is not cosmetic. If the patient believes a clinician checked the message when no meaningful review occurred, disclosure has created a false sense of security. If the organization cannot say what review means, the problem is not merely the wording of the notice; it is the workflow the notice is supposed to represent.

A disclosure can also be accurate yet insufficient. Saying that AI drafted a response does not tell the patient whether the clinician checked current guidance, considered missing information, or decided that the portal was the right channel.