AI Receptionist Script: A Practical Template and Call Examples

An AI receptionist script should make the next step clear without forcing every caller through the same conversation. Start with a short greeting, ask for the request, collect only the details needed for that request and close with an accurate account of what happens next.

For a voice assistant, the script has two parts. There are phrases the caller hears, such as the greeting. There are also instructions that guide the assistant when the caller interrupts, changes a detail or asks for something it cannot do. Writing both parts helps you move beyond a polished opening line.

Define the job before writing the greeting

Choose one workflow for the first version. For example, a repair business might want the assistant to answer approved service questions and collect callback requests. That is different from diagnosing faults, quoting a repair or booking a visit.

Write down the business name, approved information sources, details the team needs and actions actually configured. If the assistant cannot create appointments, the script should describe a request for availability, not a booking. If nobody monitors callback requests after hours, do not promise an immediate response.

Keep a simple distinction: instructions describe intended behavior; connected capabilities determine which actions are possible. The prompt-only demo guide explains why a convincing conversation does not establish that an action took place.

An AI receptionist script template

The following is an instruction template for an enquiry and callback workflow. Replace every bracketed item with verified information before testing. The brackets are editing placeholders, not automatic Burki variables. Remove any step your configured workflow cannot support.

Role: You are the AI receptionist for [business name]. Help callers with [approved enquiry types]. Identify yourself as an AI assistant in the greeting.

Greeting: “Hello, you’ve reached [business name]. I’m the AI assistant. What can I help you with?”

Conversation: Keep responses short. Ask one question at a time. Use information the caller has already provided rather than requesting it again. If the request is unclear, ask a brief clarifying question.

Business answers: Use [approved source] for factual answers. If it does not establish an answer, explain that you cannot confirm it. Do not invent prices, availability or policies.

Callback requests: If the caller wants follow-up, collect [necessary details] through the configured capture workflow. Confirm important details, including any corrections. Do not promise a response time unless [approved policy] supports it.

Actions: Use only configured, permitted actions. Describe an action as completed only when its result confirms completion. If it fails or is unavailable, explain that and offer [verified alternative].

Human requests: Follow [configured handoff process]. If a transfer is unavailable, explain the actual alternative without claiming someone has accepted the call.

Closing: Briefly confirm the request and what actually happened. Distinguish a saved request, a pending next step and a confirmed action.

This structure is a starting point, not a guarantee that the assistant will follow every instruction. Test it against real variations of the task. LiveKit's prompting guide similarly separates identity, goals, tools and limits, and recommends concise spoken responses. Keep the detailed business workflow specific to your own operation.

Greeting examples that leave room for the caller

A greeting should identify the business and make it easy to begin. Avoid announcing a long list of capabilities before the caller can speak.

Do not use a permanent after-hours greeting on a number that also answers during office hours. Likewise, “I can book that for you” belongs only in a workflow with a working booking action. A short accurate greeting is more useful than an impressive list of unconfigured services.

Example: the caller changes their request

Consider a fictional call to the repair business:

Caller: I’d like someone to look at my washing machine tomorrow.

Assistant: I can take a request for the team to review. What problem are you having with it?

Caller: It won’t drain. Actually, Friday would be better than tomorrow.

Assistant: Friday is your preferred day. What name should I put on the request?

The important behavior is retaining Friday as a preference. It is not a confirmed appointment. The assistant should not restart the entire intake simply because the caller corrected one detail.

After the configured workflow confirms that the request was saved, an appropriate closing could be: “Your request is saved with Friday as your preferred day. A visit has not been booked.” Add any follow-up expectation only if the business has supplied one. If saving fails, the closing must change to explain that failure and the available alternative.

Give uncertain answers a useful next step

“I don’t know” may be honest but still leave the caller stuck. Write a bounded alternative that the business can actually provide.

For an unsupported quote, a useful response might be: “I can’t confirm the repair cost from the information available. Would you like to leave the details for the team to review?” For an unclear reference number, ask the caller to repeat that number rather than guessing from the surrounding conversation.

Avoid telling the assistant to “always resolve the issue” or “never disappoint the customer.” Those instructions provide little guidance when a required action is unavailable. Describe the specific fallback instead, including when to stop asking questions because the caller declines to continue.

Apply and test the script in Burki

Burki provides editable assistant instructions. Adapt the template to the saved configuration and review the workflow separately: a sentence in the instructions does not connect a calendar, create a transfer destination or configure saved fields. Use the first-assistant walkthrough for the creation and browser-practice steps.

Start with fictional details and test an ordinary enquiry, a corrected date, an unknown price, a human request and a caller who declines to provide contact information. Listen for long responses, repeated questions and claims that exceed the available action results. Browser practice can help assess the conversation; telephone behavior needs testing on its intended route.

Change the instruction responsible for a failure and repeat that scenario. Keep a few successful examples too, so a correction does not quietly damage another part of the conversation. The call quality scorecard can help you record those judgments consistently.

For an AI receptionist, a useful script is one the team can explain and maintain. Keep the greeting brief, the questions relevant and the closing tied to evidence. Expand the workflow only when the assistant can handle the existing task clearly.