What the voice agent does, and what it does not
You'll be able to: describe the voice agent's scope to staff and callers, and confirm it is set to that scope.
The Practice plan includes an AI voice agent in English and Spanish, scoped to scheduling and logistics. It answers the calls your team cannot pick up: after hours, at lunch, during a session or a procedure, when two lines ring at once. You choose the rule: after hours, overflow, or every call.
It does:
- Open every call with its AI disclosure (Lesson 3.2), as it is set up for TheraCRM practices, and let the caller choose English or Spanish, then continue in that language.
- Collect scheduling and logistics only: the caller's name, callback number, the kind of visit (new patient, reschedule, a question for the office) and their availability. It does not ask for a reason for calling beyond that.
- Book into your live calendar, with no double-booking, or take a request for a callback.
- Answer logistics: hours, the address and parking, the language options, and whether the practice accepts a plan by name, from the list the practice approved. It says the office verifies benefits.
- Recognize the urgent triggers and hand off (Lesson 3.3).
- Attach the call, the transcript and the recording to the contact, and create a pipeline card.
It does not:
- Ask about or discuss health details: symptoms, history, diagnoses, medications, or why someone is seeking care beyond the kind of visit.
- Give clinical advice, say whether a treatment is right for someone, or describe expected results.
- Say a visit or treatment is covered.
- Quote prices the practice has not approved.
- Say it is a person, or use a human name that suggests it is one.
- Assess risk. A crisis phrase goes to the crisis path, every time.
- Argue with a caller who wants a person.
Why so narrow? Two reasons. First, a narrow agent is easier to supervise and less likely to say something a clinician would not. Second, the AI services behind voice agents have no published BAA status, so TheraCRM keeps health details away from the agent on every seat, and the HIPAA add-on (Module 4) keeps it that way by design. The agent schedules; people handle care.
Callers will push. "Is the filler going to hurt?" "Should I come in or go to urgent care?" "Is this normal after a root canal?" The agent says it cannot answer that, offers to book or to have a person call back, and, for anything that sounds urgent, follows the urgent path.
How it differs by practice type:
| Practice type | The agent books | It routes to a person |
|---|---|---|
| Therapy | Consults and first sessions | Any clinical question; any crisis phrase goes to the crisis path, built and tested before go-live |
| Med spa | Consults | Clinical and contraindication questions; any problem after a treatment |
| Dental | New-patient and recall visits | Pain, a broken or knocked-out tooth, swelling |
| Medical | Referral and appointment logistics | Anything that sounds like a medical emergency gets the 911 line |
Staff need to know what the agent did before they call someone back. Every call lands in the unified inbox like any other channel, with the transcript and recording on the contact.

Do it in TheraCRM
- Open the voice agent settings with the owner and read the instructions written at guided setup.
- Confirm the questions it asks are logistics only: name, callback number, kind of visit, availability.
- Confirm the list of accepted plans and approved prices it may give, or remove them.
- Place a test call in each language and ask a clinical question. Confirm it declines and offers to book or a callback.
- Check the month's AI voice minutes against your plan's allowance.
Ask the Guide
Paste this into the AI guide, without patient names or health details:
"List ten things callers to a [practice type] might ask a scheduling-only voice agent that it should decline, and a short, kind decline line for each in English and Spanish."
Next step: Hear the English and Spanish voice agent on a short TheraCRM demo.
From TheraCRM.pro