Setting expectations early — with your team and your patients — prevents most "the AI did something weird" complaints. Here's the honest list.
What the AI does well
| Task | How it works |
|---|---|
| Book new appointments | Asks for the service, preferred day/time, contact info; checks live calendar availability; confirms by SMS. |
| Reschedule | Looks up the existing appointment by phone or name, offers new slots, sends an updated confirmation. |
| Cancel | Verifies the caller, removes the appointment, optionally collects a reason. |
| Take messages | Records voicemail with a transcript and AI summary delivered to your inbox. |
| Answer FAQs | Pulls from the knowledge you added in Settings → AI Voice → Train (hours, insurance, parking, etc.). |
| Recognize returning patients | Matches caller ID to your patient list and greets them by name when found. |
| Transfer to a human | Forwards to a staff line based on your transfer rules (e.g., billing → office manager). |
| Handle Spanish + English | If you've enabled multi-language and chose a bilingual voice, the AI detects the caller's language and matches it. See Spanish & multi-lingual calls. |
| Follow custom call scripts | Branching dialogues you build in Settings → AI Voice → Call Scripts. |
What the AI won't do
| Task | Why |
|---|---|
| Give medical, legal, or financial advice | Liability. AI says "your provider can answer that — would you like me to take a message?" |
| Diagnose a symptom | Same reason. It will offer to schedule an appointment or transfer to a clinician. |
| Quote prices it wasn't given | If you didn't list a price in services or FAQs, the AI won't guess. It'll offer to text a quote or transfer. |
| Take credit card numbers over the phone | PCI compliance. AI can text a secure payment link instead. |
| Promise something outside your hours or policy | Won't book outside business hours; won't override appointment-type rules. |
| Pretend to be human if asked directly | If a caller asks "are you a real person?" the AI discloses it's an AI assistant. |
| Continue after a clear opt-out | "Take me off your list" or "stop calling" ends outbound campaigns immediately. |
| Override a staff override | Anything a staff member manually adjusts in the dashboard wins. |
Gray-area behavior you can customize
These defaults are sensible but tunable in Settings → AI Voice → Personality or Call Scripts:
- Should the AI offer same-day appointments? Default yes if slots exist. Turn off if you prefer triage.
- Should the AI mention specific providers by name? Default yes if listed. Some practices prefer to assign behind the scenes.
- Should the AI offer the cheapest option first? Default no — it offers the soonest. Configurable per service.
- Should the AI confirm insurance coverage? Default: it lists accepted plans from your FAQ but never says "you're covered" — only "we accept that plan."
When the AI escalates
Built-in triggers that hand off to a human (or voicemail when none available):
- Caller says "speak to a person" / "talk to someone" / "agent".
- Caller is upset (sentiment detected as frustrated for two turns).
- Caller asks something clearly outside the AI's scope (legal threat, emergency, billing dispute).
- The AI fails to understand twice in a row.
- Caller requests a transfer by name (e.g., "Can I speak to Dr. Park?").
You set the destination in Settings → Phone → Transfer Rules. If no staff line is configured, the call goes to voicemail with a summary delivered to your inbox.
Reality check before going live
Run all six built-in call scripts before you start advertising the number. See Test your AI with call scripts. Patterns the AI handles smoothly in scripts will be the same patterns it handles smoothly with real callers.