Before the first AI call: what patients should know, and what consent should say

Nobody joins a practice for its consent forms. But the first time a patient realises the friendly receptionist on the phone was an AI, the practice’s reputation depends on whether they were told, and how. Here is what we tell patients at Genesis Health, and what our consents actually say.

Say what it is, plainly

The AI introduces itself as the practice’s AI receptionist, in the first sentence. It says what it can help with (booking, questions about the practice, forms) and that a person can be reached at any time by asking. The same disclosure is on the website and in the portal. Patients do not mind talking to an AI; they mind being fooled.

Texting is a consent, not a default

Reminders and check-ins by text need the patient’s agreement, recorded with the date and the channel it was given on. The consent is specific: appointment texts, follow-up check-ins, or both. Anything clinical goes through the secure portal, and the patient gets a text with a link rather than the details themselves. A patient can withdraw with one reply, and the practice sees the change immediately.

Recording a visit is its own decision

Consent to have the AI scribe listen during a visit is separate from everything else, explained by a person, and stored with the note it produced. It covers what is recorded, who reviews the draft, how long audio is kept, and how to withdraw. If a patient says no, the visit is not recorded. There is no version of this where the practitioner “just knows” the patient is fine with it.

Consent is not a signature on file. It is a patient being able to explain, in their own words, what they agreed to.

Log everything the AI does

Every message sent, call answered, appointment booked and hand-off to a person is written to the record with a timestamp and the reason. The practice reviews this in one place. A patient who asks “what has the AI done in my name?” can be given a straight answer. Vendors that cannot produce this log are asking for trust they have not earned.

The escalation rule, in writing

Patients should know the one rule that never changes: clinical questions go to their practitioner. Write it down. Put it in the consent, on the website and in the AI’s own script. It is the single sentence that makes the rest of the system acceptable to a careful patient, and it happens to be the sentence that makes the system safe.

A short checklist

Does the AI say it is an AI? Can a patient reach a person by asking? Are texting and recording separate, dated consents? Can either be withdrawn in one step? Is every AI action logged and reviewable? Is the clinical-escalation rule written where patients can read it? Six questions. If any answer is no, fix that before the first call.