Blog · · 6 min read

AI Receptionist for Dental Clinics: Phone Booking, Changes and Cancellations

How to let AI answer a dental booking line: checking openings, booking, rescheduling and cancelling, verifying patients, and handing urgent calls to staff.

Key points

  • Dental booking calls fall into a small set of intents (check availability, book, reschedule, cancel), which makes them a good fit for a voice AI.
  • The AI checks open slots through the booking system's API during the conversation and reads the details back for a yes before it creates, changes or cancels anything.
  • Urgent symptoms, clinical questions and anything the AI can't verify should be handed off to front-desk staff, not answered.

Yes, an AI can answer a dental clinic's booking line. Most calls want one of four things: an open slot, a new appointment, a different time, or a cancellation. The AI looks up availability in the booking system during the conversation, reads back what it's about to do, and only writes after the caller says yes. Clinical questions and urgent pain go to a person.

Why dental booking calls suit a voice AI

The call types are few and the steps are predictable. That's exactly what you want to automate.

Caller wants to The AI Booking system operation
Check availability Asks for a preferred day and time of day, offers open slots Find free slots
Book Collects name, callback number and preferred time, then books Find free slots, create appointment
Reschedule Verifies the patient, confirms the current and new times List appointments, change appointment
Cancel Verifies the patient, confirms which appointment List appointments, cancel appointment
Ask a common question Answers hours, closures, parking and so on Read clinic info

The front desk at a dental practice is usually also checking patients in and out. Every phone call pulls someone away from the person standing in front of them. Letting an AI take the routine booking calls means staff only pick up the calls that actually need a human.

The call flow

Run every booking call through the same five steps:

  1. Greet and ask what they need. "Thanks for calling Bright Smile Dental. Are you calling to make or change an appointment?"
  2. Identify the patient. New or existing? For anything that looks up or changes an existing appointment, verify identity first.
  3. Find a time. Ask for a day and a rough time of day, then offer two or three options. Long lists don't work over the phone; people can't hold them in their head.
  4. Read back and get a yes. "That's Friday, November 28 at 3 p.m. for John Smith. Shall I book it?" Only create the appointment after a clear yes.
  5. Close. Repeat the time briefly and mention anything they need to bring.

Rescheduling and cancelling use the same steps 2 and 4. Never let the AI change a booking without reading it back first; we go deeper on this in read-back confirmation.

Verifying who's calling

Before the AI looks up, changes or cancels an appointment, confirm the caller is the patient (or a family member handling it).

  • Verify with the patient card number. Ask for it before looking up, changing or cancelling an appointment.
  • Use the phone number only to find the patient. A matching caller ID doesn't prove who is calling: family members and others call from the same phone. Use the number to find the record, not to verify.
  • Patients without a card number. Someone who first booked by phone may not have a card number on file yet. They can't be verified on the phone, so don't read out their booking; tell them staff will call back.

Don't ask for date of birth. Fewer questions make the call shorter and leave less personal data in the record.

Don't let the model do the matching. Have the booking system compare the values and return only "verified" or "not verified". The model never needs to see the stored values, which keeps sensitive data out of prompts and logs.

What the AI should not answer

Some calls should never be handled by the AI. Put these in your handoff conditions:

  • Urgent symptoms: severe pain, swelling, bleeding that won't stop, a knocked-out tooth
  • Questions about treatment, cost estimates or medication
  • The caller asks for a person
  • The patient can't be verified

When one of these comes up, the AI ends its part and the call goes back to your phone flow, which can ring the front desk. What to pass along at that moment is covered in designing AI-to-human handoff.

Dental calls also touch health information. Collect only what you need to book, and don't let the AI ask about symptoms in detail. If you operate in the US, check how HIPAA applies to your call handling and vendors; in Japan, medical history is "special care-required personal information" under the APPI. More in privacy in medical phone calls.

Connecting to the booking system

To check availability and book during the conversation, the AI needs to call the booking system's API mid-conversation. How you connect depends on where the clinic keeps its schedule:

Where the schedule lives How to connect
A booking system with an API Expose its API to the AI as in-call tools
A calendar such as Google Calendar Have your own server call the calendar API to find slots and add events
A paper book or a system without an API Don't book on the call; send the collected details to the front desk afterwards

Whichever you choose, separate read operations (finding slots) from write operations (create, change, cancel), and require confirmation before writes. Without an API, receive the name, callback number and preferred time through the post-call webhook, and have staff call back to book. See function calling in live calls for how to design the tools themselves.

Doing this with voicast

In voicast you describe the receptionist in a agent settings (an agent): role, greeting, fields to collect, handoff conditions and FAQ. The booking system is reached through in-call tools (config.tools). Each tool has a name, a description of when to use it, a URL and its parameters. When the AI decides to use a tool, voicast sends a signed POST to that URL. Your endpoint calls your booking system or calendar API and returns JSON.

ts
await fetch('https://api.voicast.jp/v1/agents', {
  method: 'POST',
  headers: {
    authorization: `Bearer ${process.env.VOICAST_API_KEY}`,
    'content-type': 'application/json',
  },
  body: JSON.stringify({
    name: 'Dental booking line',
    config: {
      greeting: 'Thanks for calling Bright Smile Dental. Are you calling about an appointment?',
      persona: 'A calm, friendly dental receptionist. Short, polite sentences. Never discuss treatment or prices.',
      fields: [
        { key: 'name', label: 'Name', required: true },
        { key: 'purpose', label: 'Reason for the call', required: true },
      ],
      handoff_when: [
        'Urgent symptoms such as severe pain, swelling or bleeding',
        'Questions about treatment, cost or medication',
        'The caller asks to speak to a person',
        'The patient could not be verified in the booking system',
      ],
      closing: 'You are all set. We look forward to seeing you.',
      domain: 'dental clinic',
      vocabulary: ['cleaning', 'filling', 'crown', 'check-up', 'dentures'],
      tools: [
        {
          name: 'find_free_slots',
          description: 'Find open slots. Use when the caller names a preferred day.',
          url: 'https://clinic.example.com/voicast/tools',
          parameters: {
            properties: {
              date: { type: 'string', description: 'Preferred day (YYYY-MM-DD)' },
              menu: { type: 'string', enum: ['check-up', 'treatment', 'cleaning'] },
            },
            required: ['date'],
          },
        },
        {
          name: 'verify_patient',
          description: 'Verify an existing patient. Ask for the patient number, then use this before looking up, changing or cancelling an appointment.',
          url: 'https://clinic.example.com/voicast/tools',
          parameters: {
            properties: {
              name: { type: 'string' },
              card_number: { type: 'string', description: 'Patient number' },
            },
            required: ['name', 'card_number'],
          },
        },
        {
          name: 'create_booking',
          description: 'Book an appointment. Use after offering slots and the caller picks one.',
          url: 'https://clinic.example.com/voicast/tools',
          confirm: true,
          parameters: {
            properties: {
              start: { type: 'string', description: 'Start time (ISO 8601)' },
              name: { type: 'string' },
              phone: { type: 'string' },
            },
            required: ['start', 'name', 'phone'],
          },
        },
        {
          name: 'cancel_booking',
          description: 'Cancel an appointment. Use after verifying the patient and confirming which appointment.',
          url: 'https://clinic.example.com/voicast/tools',
          confirm: true,
          parameters: {
            properties: { booking_id: { type: 'string' } },
            required: ['booking_id'],
          },
        },
      ],
    },
  }),
});

Tools that book or cancel get confirm: true: the AI reads the time and name back and only uses the tool after the caller says yes. The verification tool checks the patient number the caller gave on your server and returns only the result (verified or not, plus the appointment ID and time), never the stored values. One URL can receive every tool and route by the tool name.

Your endpoint checks the signature before calling the booking system. Run each tool_call_id only once, so a repeated request never creates a second appointment. Reply within 8 seconds. If the booking system is down, the AI apologizes and offers a callback from staff.

When the conversation ends, the collected fields and a summary arrive by webhook. From there you can post to the front desk's Slack or chat, or log the call in the booking system. Built-in integrations such as Google Calendar are in preparation. More sample setups are on the examples section.

FAQ

Can the AI book new patients too?

Yes. It collects a name and callback number, finds a slot, reads the details back and books after the caller confirms. Your closing line can remind them what to bring.

What happens when someone calls with a dental emergency?

If urgent symptoms are in your handoff conditions, the AI doesn't try to answer; it hands off. The call returns to your phone flow, which can ring the front desk or, after hours, play whatever emergency guidance the clinic uses.

Could a caller change someone else's appointment?

Not if you verify before every lookup or change. Verify with the patient number (a matching caller ID alone isn't enough), and do the comparison in the booking system so the AI only receives the result.

Related posts