---
name: ai-voice-receptionist
description: Design, configure and test an AI phone receptionist for a business (answer, understand intent, book, transfer and log a summary in the CRM). Use when the user wants a virtual receptionist, a voice assistant on their number, or to automate phone intake.
---

# AI Voice Receptionist

## Goal
A voice that answers every call, handles the simple things with verified information, books appointments and passes everything else to a person. Nothing made up.

## Step 1. Knowledge base (nothing gets built without it)
Ask the user and write down:
- Business, address, service area, hours and holidays.
- Services: what each includes and what they do NOT do.
- Policies: cancellations, payment methods, appointment requirements.
- The 15 questions callers ask most, with the owner's exact answer.
- Pricing: only if the owner wants it said on the phone; otherwise the answer is "a team member will confirm that".

## Step 2. Script
1. Greeting with the business name and a clear disclosure that this is an AI assistant.
2. Open question: "How can I help?"
3. Qualifying questions (3 max): what they need, when, where.
4. Action: answer from the base, book, take a message or transfer.
5. Close: confirm what was agreed (day, time, name, phone) and say goodbye.

## Step 3. Guardrails (put them in the system prompt)
- Never invent prices, diagnoses, availability or promises.
- If it is not in the base: "A team member will confirm that for you" and take a message.
- Transfer immediately: emergencies, complaints, legal or medical topics, or when the caller asks for a person.
- Never collect sensitive data (cards, passwords) by phone.
- Disclose call recording as local law requires.

## Step 4. Connections
- Voice platform: whatever the user runs (for example GoHighLevel's voice AI or another voice platform).
- Calendar: check real availability before offering times.
- CRM: on hang-up, create or update the contact with summary, intent, outcome and recording.
- Team alert for messages, failed transfers and new bookings.

## Step 5. Test before going live
Make at least these calls and note what happened:
- Normal customer who books.
- Question that is not in the base.
- Upset caller who asks for a person.
- Call in another language or with heavy background noise.
- Someone asking about prices.
Fix the prompt and repeat until all of them pass.

## Operating rules
- Stays in test mode until the owner approves in writing.
- Weekly review: listen to a sample of calls and update the base.
- Deliver: final script, knowledge base, test list with results, open items.