# Plan an AI receptionist that answers routine calls and hands off the rest
By The Workshaped Lab · Tier: Agentic flow · Industry: Healthcare, Construction & Trades, Agencies & Professional Services · Function: Customer Service · Mode: Human approval · Autonomy: Act with approval · Prompt tested: 2026-10-14 (gpt-5-mini, synthetic sample, checked by hand)
Provenance: Idea, prompt tested on synthetic data. Tools: A voice-agent service (e.g. Retell, Vapi, your phone provider's AI), Booking system, Any chatbot for the call script

Front desks spend most of the day on the same five questions. A voice agent can answer those and take booking requests, as long as it knows exactly when to hand the call to a person. Most of the work is the call script and the handoff rules, which you can draft today.

## The problem
Clinics, practices and service businesses miss calls at peak times and lose bookings, while staff repeat opening hours, prices and directions all day. Voice agents can help, but a badly scoped one frustrates callers or gives answers it shouldn't (medical advice, prices it made up).

## Why it works
The safe version is narrow: a fixed list of questions it may answer from a fact sheet you wrote, a booking request it collects but a person confirms, and clear triggers to transfer (anything clinical, angry callers, unclear requests). Writing that script and fact sheet is the part a chatbot can help with today; any voice-agent tool can then use it.

## Who it's for
Clinics, dental and physio practices, salons, law and accounting offices, trades with a busy phone line.

## Step by step
1. List the 10 most common calls from a week of call notes.
2. Write a fact sheet: hours, address, parking, services, prices you publish, cancellation policy.
3. Write the call script: greeting, the questions it may answer, how it collects a booking request (name, number, reason, preferred times).
4. Write handoff rules: always transfer anything clinical, urgent, a complaint or anything not in the fact sheet.
5. Test the script by role-playing 20 calls in a chatbot, then load it into a voice-agent tool on a test number.
6. Go live for overflow calls only, and review recordings weekly.

## Pitfalls to avoid
- Healthcare: it must never give medical advice or triage; transfer and give the emergency number.
- Recording and privacy rules: tell callers they're speaking to an AI and if calls are recorded; check HIPAA/BAA terms with the vendor before handling patient details.
- Bookings must be requests a person confirms until you trust it.
- Costs are per minute (typically several cents a minute); start with overflow only.

## Free kickstart prompt (paste into Claude, ChatGPT or Grok)
```
I run a [TYPE OF BUSINESS] and want an AI voice receptionist for routine calls. Here are our facts (hours, address, services, published prices, policies):
[PASTE FACTS]
And the most common reasons people call:
[PASTE LIST]
Write:
1. A call script: greeting that says it's an AI assistant, how it answers each common question using ONLY my facts, and how it collects a booking request (name, phone, reason, 2 preferred times) without confirming a time.
2. Handoff rules: an exact list of situations where it must transfer to a person or give the emergency number, including anything medical, urgent, a complaint, or not covered by my facts.
3. 10 test calls (caller lines only), including 3 tricky ones, so I can role-play it before going live.
Never make up prices, availability or advice that isn't in my facts.
```

Web: https://workshaped.ai/use-cases/voice-receptionist/