How to Write an AI Voice Agent Script That Books Appointments

What should an AI voice agent booking script include?

A booking script for an AI voice agent needs six parts: a greeting that names your business, a reason-for-call question, three to five qualifying questions, a short list of facts the agent can answer, rules for offering times, and a handoff rule for anything it can't handle. Write each part in plain spoken English, the way your best receptionist talks.

Most people write their first script like a website. It comes out long, formal and full of features. Then they hear it on a real call and it sounds like a robot reading a brochure. A phone script is shorter than you think. Every line the agent says should either move the caller toward a booked time or answer something they asked.

Here is the full structure I'd start with for a service business:

1. Greeting. Business name, the agent's name, one short question.
2. Reason for the call. Let the caller say what they need in their own words.
3. Qualifying questions. The three to five things you need to know before you'd send someone out or take a meeting.
4. Facts the agent may share. Hours, service area, price ranges you're happy to quote, what to bring.
5. Booking rules. Which calendar, how far out, how long the slot is, and what to confirm before hanging up.
6. Handoff rules. When to transfer to a person, take a message or end the call politely.

How do you write the opening line for an AI booking agent?

Open with your business name, a first name for the agent and one easy question, all in under ten seconds. Something like: "Thanks for calling Lakeside Plumbing, this is Ava. What can I help you with today?" Short openers keep callers talking; long ones make them press zero.

For outbound calls, consent matters before wording does. In February 2024 the FCC confirmed that AI-generated voices count as "artificial" voices under the Telephone Consumer Protection Act (FCC 24-17), so the same consent, identification and opt-out rules apply as for prerecorded calls. That means the opener has to say who is calling, and the agent needs a clean way to stop calling anyone who asks. We cover the details in our guide to the FCC rules on AI calls.

Decide up front what the agent says if someone asks whether it's a real person. Have it answer honestly in one sentence and keep going. Some states already require this. Utah's 2025 update to its AI law kept the duty to disclose when a consumer directly asks (Future of Privacy Forum summary). A straight answer costs you nothing and keeps the caller's trust.

What questions should the agent ask to qualify a caller?

Ask only the questions you'd need answered before you'd spend time on the job: what they need, where they are, how soon, and one question that filters out bad fits. Three to five questions is enough. Every extra question is another chance for the caller to hang up before they book.

Here's an example set for a home service business. Swap in your own, but keep the same idea: every question has a reason and a rule for what happens next.

Write the questions the way you'd say them out loud. "Can I get your zip code?" works on the phone. "Please provide the service location postal code" does not.

How should an AI agent offer appointment times?

Offer two specific times, the sooner the better, instead of asking "when works for you?" Give a choice like "I have today at 4pm or tomorrow at 10am, which is better?" Two options make the decision easy, and near-term slots get far fewer no-shows than bookings a week out.

Our own numbers back this up. In our own calendar data, over 45 days, calls our AI agents booked went 10 showed and 14 no-shows. Most of the no-shows were booked 2 to 5 days out. Most of the people who showed were booked within 24 hours. So I tell the agent to lead with the earliest open slots and only go further out if the caller asks.

Time zones caused one of those no-shows. A caller in another time zone booked a slot that looked fine to them and was wrong on our end. Since then, every booking script I write has the agent confirm the time zone out loud for any caller outside your area.

Before the agent ends the call, it should read back four things:

• The day, date and time, with the time zone
• The address or meeting link
• What the caller should have ready (photos of the problem, documents, the decision maker)
• That a confirmation text is on its way, and to reply if anything changes

What should the agent do when it can't answer a question?

Give the agent a short list of handoff rules: transfer live during business hours, take a detailed message after hours, and never guess at prices, diagnoses or legal answers it wasn't given. A clear "I'll have someone call you back about that" beats a confident wrong answer every time.

I write the handoff rules as a list of triggers. These cover most service businesses:

• The caller asks for a person twice: transfer, or take a message if no one is free.
• The caller is upset or mentions a complaint: apologize once, take the details and flag it for the owner the same day.
• The question is about a price you haven't approved for the agent to quote: give the range you did approve, or offer a callback.
• Anything about safety, like a gas smell or an electrical fire: tell them to call 911 or the utility first.
• The caller says stop calling: confirm, end the call and mark the contact so no one calls again.

Also give the agent a list of things it can always share: hours, service area, whether you charge a trip fee, how long a typical visit takes. The more of these you write down, the fewer handoffs you'll get.

How do you test an AI agent script before it goes live?

Call the agent yourself at least 20 times with different scenarios, listen to every recording, and fix the script after each round. Test easy bookings, wrong-area callers, people who want a human, people who interrupt, and someone in a different time zone. Go live only when it handles all of them without help.

1. Write ten test scenarios from real calls you've had, including the hard ones.
2. Have two or three different people call, because accents and speaking speed change how the agent hears things.
3. Check that every test booking actually lands on the right calendar with the right time zone and contact details.
4. Listen for lines that sound stiff, and rewrite them shorter.
5. After launch, review ten real calls a week for the first month. Fix the script based on what callers actually say.

Speed is the other half of this. A Harvard Business Review study of 1.25 million sales leads found that firms that tried to reach a lead within an hour were nearly seven times as likely to qualify them as firms that waited even an hour longer. An agent that answers on the first ring and books on the spot handles that for you. Our post on why leads go cold goes deeper on the numbers.

Your next step: pull up your last ten booked calls, write down the questions you asked on each one, and use that list as your first draft. If you'd rather have us build and test the agent with you, book a call.