AI Receptionist Scripts and Prompt Templates (Copy-Paste, 2026)

What an "AI Receptionist Script" Actually Is

When people search for an ai receptionist script, they usually have one of two things in mind, and conflating them is the fastest way to build an agent that sounds robotic.

The first is the system prompt — the standing instructions the language model reads on every call. This is the brain. It defines who the agent is, what it knows, what it is allowed to do, and how it behaves when a caller goes off-script.

The second is the conversation script — the actual words: the opening greeting, the booking flow, the answers to the five questions every caller asks. This is the mouth.

A good voice agent needs both, and they work differently. The system prompt is rules and knowledge. The script is example phrasing the model imitates. If you only write a script, the agent breaks the moment a caller says something you didn't anticipate. If you only write a system prompt with no example phrasing, the agent sounds generic and stilted. You want both, layered.

Everything below is copy-paste. Swap the bracketed [PLACEHOLDERS] for the client's details and you have a deployable agent. These drop into the prompt field on Retell, Vapi, Synthflow, or Bland with no changes to the structure — only the platform's calendar/tool wiring differs, which we cover at the end.

The System Prompt Scaffold (Use This Every Time)

Most beginner prompts fail for the same reason: they're a wall of text with no structure, so the model can't tell a rule from a fact from a suggestion. The fix is a five-block scaffold — Identity, Scope, Knowledge, Behavior, Booking — kept under roughly 500 words total. Long prompts add latency and, counterintuitively, increase hallucination because the model loses the thread.

Here is the scaffold. This is the single most reusable asset in this post.

# IDENTITY
You are the front-desk assistant for [BUSINESS NAME], a [BUSINESS TYPE]
in [CITY, STATE]. You are warm, brief, and efficient. If a caller asks
whether you are a person, say: "I'm the AI assistant for [BUSINESS NAME] —
happy to help." Never pretend to be human.

# SCOPE
You CAN: answer questions about hours, location, services, and pricing;
book, reschedule, and cancel appointments; take messages.
You CANNOT and WILL NOT: give [professional/medical/legal] advice, quote
prices that are not listed below, or make promises about availability you
have not confirmed with the calendar tool.

# KNOWLEDGE (the only facts you may state)
- Hours: [Mon-Fri 8am-5pm; closed weekends]
- Location: [123 Main St; free parking in the rear lot]
- Services & prices: [list each service and its price or "call for quote"]
- Insurance/payment: [what you accept; what you do not]
- New customers: [intake steps, deposit, what to bring]

# BEHAVIOR RULES
- Keep every reply under two sentences. Ask one question at a time.
- If you do not know something, say so and offer to take a message.
  NEVER guess a price, a policy, or an open time slot.
- If the caller is angry, or describes [an emergency: severe pain,
  bleeding, a safety threat], say you'll connect them to a person and
  trigger a transfer to [TRANSFER NUMBER].
- Confirm names and phone numbers by reading them back digit by digit.

# BOOKING INSTRUCTION
When a caller wants an appointment:
1. Collect full name, callback number, and preferred day/time.
2. Call the availability tool — never invent open slots.
3. Offer up to two real openings.
4. On confirmation, call the booking tool and read back the details.

The two load-bearing lines in that whole scaffold are "NEVER guess a price" and "never invent open slots." Those two rules prevent about 90% of the embarrassing failures that lose a client in week one. An agent that says "I'm not sure, let me take a message" sounds competent. An agent that confidently quotes a wrong price sounds like a liability.

The Greeting Script

The opening line does three jobs at once: it identifies the business, it discloses that the caller is talking to an AI, and it invites a request — all in one breath so it never feels like a robot reading a disclaimer.

Thanks for calling [BUSINESS NAME], this is the front-desk assistant.
How can I help you today?

That phrasing satisfies the soft AI-disclosure expectation in states like California for inbound calls, because "front-desk assistant" plus the identity line in the prompt makes the automation clear without a clunky legal warning. (For outbound calling the rules are much stricter — see our note on TCPA compliance for AI voice agents before you ever dial out.)

Set interruption sensitivity to medium so the agent stops talking the instant a caller jumps in. An agent that talks over people reads as robotic no matter how good the voice model is.

The Booking Flow Script

This is the part that turns a $99/month toy into a $297–$497/month service. Give the model example phrasing for the happy path and the friction paths, then let the system prompt's booking instruction govern the tool calls.

CALLER: I'd like to book an appointment.
AGENT:  Happy to help. Can I get your full name?
CALLER: [name]
AGENT:  Thanks, [first name]. What day works best for you?
CALLER: Thursday morning.
AGENT:  Let me check Thursday morning... I have 9:15 or 10:30 open.
        Which works better?
CALLER: 10:30.
AGENT:  Got it. And the best number to reach you?
CALLER: [number]
AGENT:  Let me read that back — [number, digit by digit]. Correct?
CALLER: Yes.
AGENT:  Perfect. You're booked for Thursday at 10:30 for [service].
        Anything else I can help with?

-- If no slot fits --
AGENT:  I don't have that exact time, but I have [alt 1] or [alt 2].
        Would either work, or should I have someone call you back?

-- If the request is out of scope --
AGENT:  That's a great question for [the doctor / our specialist].
        I'll take your name and number and have them call you back today.

Notice the agent never says "yes, that time is open" before the availability tool confirms it. The script models the behavior the system prompt requires. That redundancy is deliberate — it's why the agent holds up under pressure.

The FAQ Script

Roughly five questions cover the large majority of inbound calls to a local service business: hours, location, price, "do you take my insurance/payment," and "are you open right now." Pre-write these so the model has clean phrasing to imitate, and keep every answer to one or two sentences.

Q: What are your hours?
A: We're open [Mon-Fri, 8 to 5], closed weekends. Want me to book you a time?

Q: How much does [service] cost?
A: [Service] is [price]. If you'd like, I can get you on the schedule.

Q: Where are you located?
A: We're at [address], with free parking in the rear lot.

Q: Do you take [insurance/payment]?
A: [We accept X and Y; we don't currently take Z]. Want me to set something up?

Q: Are you open right now?
A: [Yes, until 5pm today / No, we open at 8am tomorrow]. I can still book you now.

Every FAQ answer ends by steering back toward a booking. A receptionist that answers questions is worth a little; one that consistently converts the question into an appointment is worth the monthly fee.

Adapting the Scaffold Per Niche

The five blocks never change. Only the Knowledge block and the emergency rule do. A few examples of what moves:

Niche	Emergency / escalation trigger	Watch-out
Dental / medical	Severe pain, bleeding, trouble breathing → transfer	HIPAA: don't store PHI loosely — see the HIPAA-compliant guide
Law firms	"Statute of limitations," arrest, court date → transfer	Never give legal advice or quote case outcomes
Home services / auto repair	Gas leak, no heat in winter, smoke → urgent dispatch	Quote ranges only; confirm trip-charge policy
Med spa / salons	Adverse reaction post-treatment → transfer	Deposit/cancellation policy stated up front

When in doubt, tighten scope rather than widen it. A narrow agent that books cleanly and hands off everything else beats a know-it-all that improvises.

Where Scripts Meet the Platform

The scaffold above is platform-agnostic, but the booking and availability steps are wired differently depending on where you build:

• Retell (~$0.07/min) and Bland (~$0.09/min) bundle the stack; you attach a calendar tool in the dashboard. Start with the Retell tutorial.
• Vapi (~$0.05/min + your own LLM/voice costs) is modular — you wire tools via the API. See the Vapi tutorial.
• Synthflow (~$29–$500/mo tiers) is the most no-code; the scaffold pastes straight into its prompt field.

Full per-platform numbers live in our AI receptionist cost breakdown, and you can compare options on the tools and compare pages.

A Realistic Note on Iteration

Your first prompt is never the deployed prompt. Plan to call the number yourself and run five scenarios before go-live: a clean booking, a pricing question, an out-of-scope question, an interruption mid-sentence, and an emergency phrase. You're listening for sub-second latency, zero invented facts, and a clean handoff when the agent hits its limit. Expect two or three rounds of edits. That's normal, not a sign the scaffold is wrong.

Get the Full Library Instead of Building From Scratch

These scaffolds are enough to ship a working agent today — copy them, fill the brackets, test, deploy. The slow part isn't the first script; it's rebuilding all of this for every new client and every new niche.

That's the gap the AI Receptionist Agency Launch System closes. The Core kit ($497) includes a Claude Prompt Library for Voice AI — 150+ prompts that extend exactly the scaffold above across niches — plus the agency playbook, a Twilio + LiveKit + Retell setup blueprint, an ROI calculator with a word-for-word sales script, client-acquisition campaigns, proposal and contract templates, a Notion command center, a recorded onboarding walkthrough, and 30 days of email support. Premium ($997) adds a 60-minute 1:1 setup call, a proposal review, and 60 days of priority support. Both carry a 60-day land your first client or full refund guarantee.

If you'd rather grab the prompt library and blueprint à la carte, they're on the shop page. Either way, copy the scaffold above first — it's the foundation everything else builds on.