An AI receptionist script is the set of instructions that tells a voice agent how to greet callers, identify intent, collect useful details, answer approved questions, take an action, and escalate when needed. A strong script defines decisions and boundaries, not a rigid line-by-line speech for every possible call.
Last updated: July 31, 2026. By the Telvana team.
What is an AI receptionist script?
An AI receptionist script is an operating playbook for phone conversations. It defines what the receptionist is responsible for, what information it can use, which questions it should ask, which actions it may take, and when a person must step in.
That is different from a traditional script that gives a receptionist exact lines to read. Callers interrupt, change topics, answer questions out of order, and ask for things the business did not anticipate. The playbook needs enough structure to produce a reliable next step without forcing every call through the same sequence.
The goal is not to predict every sentence. The goal is to make the important decisions explicit.
The eight parts every AI receptionist script needs
A practical script should cover eight parts. Each part answers a different operational question.
| Part | What it defines | Example decision |
|---|---|---|
| Role and scope | Who the receptionist represents and what it may handle | Answer routine questions and route service requests |
| Greeting | How the call begins | State the business name and invite the caller to explain |
| Intent discovery | How the receptionist identifies the reason for the call | Separate a new lead from an existing customer request |
| Information collection | Which details are required for each intent | Capture name, callback number, service need, and timing |
| Approved answers | Which business facts may be answered directly | Use the current knowledge base for hours and service areas |
| Action rules | What happens after the needed details are collected | Book, transfer, route, or take a message |
| Escalation and fallback | What happens when confidence, access, or authority runs out | Transfer to a person or capture a callback request |
| Confirmation and close | How the receptionist verifies the outcome | Repeat the booking time or promised next step |
These parts connect directly to the work a receptionist performs. Telvana's current call handling features include knowledge base answers, routing, appointment scheduling, call records, and CRM or calendar connections. The script should define how those capabilities fit the business instead of treating them as one generic flow.
AI receptionist script template you can adapt
Use this template as a starting structure. Replace every bracketed field with a business rule that your team has reviewed.
Role and purpose
You are the phone receptionist for [Business Name]. Help callers reach a useful next step. You may answer questions from the approved business knowledge base, collect details for defined call types, book approved appointments, transfer eligible calls, and take messages. Do not guess when the answer or action is not authorized.
Greeting
Thank the caller for contacting [Business Name]. Introduce yourself using the approved receptionist name. Ask an open question such as, “How can I help today?” Do not begin with a long menu unless the business has a clear reason to use one.
Intent discovery
First identify the caller's main reason for calling. Use the business's actual categories, such as:
New service request
Existing customer question
Appointment request
Billing or account question
Vendor or partner call
Urgent request that follows a defined escalation path
Other request
If the caller gives enough information in the opening sentence, do not ask the same question again.
Information collection
Collect only the details needed to complete the next step. For a new lead, that might include name, callback number, requested service, location, and timing. For an existing appointment, that might be name, appointment date, and the requested change.
Confirm important details before taking action. If the caller corrects a detail, use the corrected version and continue from there.
Approved answers
Answer only from the current knowledge base and approved business rules. Common topics may include hours, service area, appointment types, location, and basic process questions.
If a question is outside the knowledge base, do not invent an answer. Follow the configured fallback path. Telvana's FAQ describes transfer, message, and callback options for questions the receptionist cannot answer.
Action rules
For each call type, define one preferred action and one fallback action.
Book an appointment only when the requested appointment type and time are available under the approved scheduling rules.
Transfer a call only when the intent, destination, and transfer window meet the business's rules.
Route a lead to the correct person, team, or workflow based on the information collected.
Take a detailed message when booking, routing, or transfer is not available.
Tell the caller what will happen next.
Escalation and fallback
Escalate when the caller asks for a person, the request is outside the receptionist's scope, required information is unavailable, or the configured workflow says a human must decide.
If a transfer fails or nobody is available, return to the caller with the approved message path. Confirm the best callback detail and set a realistic expectation based on the business's actual process. Never promise a response time the team has not committed to.
Confirmation and close
Before ending the call, confirm the completed action and any next step. For a booking, repeat the date, time, and location. For a message, confirm who will receive it and what detail was captured. Then ask whether the caller needs help with anything else.
Scenario rules turn the script into a workflow
A script becomes useful when it maps common caller intents to outcomes. Start with the call types your business actually receives, not an imaginary list of every possible situation.
| Caller intent | Minimum useful details | Preferred action | Fallback |
|---|---|---|---|
| New lead | Name, callback number, need, location, timing | Route or book under approved rules | Capture a complete message |
| Existing customer | Name, callback number, account context, request | Route to the responsible team | Record a callback request |
| Appointment request | Name, appointment type, preferred timing | Book an available approved slot | Offer a callback or alternate path |
| General question | The exact question | Answer from the current knowledge base | Transfer or take a message |
| Urgent request | Nature of the request, location, callback number | Follow the business's approved escalation path | Use the defined backup contact or message path |
| Vendor or partner | Company, caller name, purpose | Route to the correct contact | Take a message |
Do not force every caller to answer every field in the table. Ask only what the selected path requires. A real estate team, for example, needs different questions for buyer, seller, and acquisition calls. Our real estate call flow guide shows how one industry can turn those distinctions into specific routing rules.
How to write qualification questions that feel natural
Qualification should move the call forward, not make the caller feel trapped in an intake form.
Use these rules:
Ask one question at a time.
Start broad, then ask for the detail needed to choose an action.
Reuse information the caller already provided.
Explain why a sensitive or unexpected detail is needed when the business has approved that question.
Stop collecting information once the next step is clear.
Confirm details that affect booking, routing, or follow-up.
Compare these two approaches.
A weak sequence asks for name, phone number, email, address, service, budget, and timing before acknowledging the caller's problem.
A better sequence starts with the reason for the call, then asks only for the details needed to route or book that request. It can collect contact information after the caller understands what will happen next.
Google Cloud's general conversational agent design guidance recommends starting with top-level requests, building the basic structure, and then iterating across the routes a user may take. The same principle applies here. Begin with the call paths that matter most, then add detail based on real conversations and failed test cases.
Define booking, transfers, and fallback behavior separately
Booking, transfers, and messages are different outcomes. Give each one its own rules.
Booking rules
Specify which appointment types may be booked, which calendars or resources apply, what details are required, and what to do when the requested slot is unavailable. Avoid vague instructions such as “book something that works.”
Transfer rules
Specify which intents qualify, who receives each transfer, the hours when the destination is available, and the backup behavior if the transfer cannot connect. Tell the caller what is happening before attempting the handoff.
Message and callback rules
Specify which details make a message actionable, where the message goes, and what expectation the receptionist may set. A message without the reason for the call or a usable callback detail simply moves the same problem to a person later.
This separation also makes the script easier to maintain. A change to appointment availability should not accidentally change the transfer policy.
Test your AI receptionist script before launch
A script is ready only after it has handled both normal and difficult test calls. The NIST AI Risk Management Framework places trustworthiness considerations across AI design, development, use, and evaluation. In practical terms, testing should be part of the call flow, not a one-time box checked after setup.
Run at least these scenarios using your own business rules:
A common new lead with complete information
A caller who gives details out of order
A caller who changes intent halfway through the call
A question that is not in the knowledge base
A requested appointment time that is unavailable
A transfer destination that does not answer
An after-hours call with a valid request
A caller who asks for a person immediately
A noisy or ambiguous response
A call where the receptionist must correct a misunderstood detail
For every test, verify the greeting, captured details, chosen action, confirmation, and record of the conversation. Google Cloud's guidance also recommends defined test cases so changes do not reintroduce old failures.
Keep a simple test log with the scenario, expected outcome, actual outcome, correction, and retest result. The goal is not a perfect performance in one clean demo. The goal is predictable behavior across the calls your business receives.
Common AI receptionist script mistakes
Writing a monologue instead of a conversation
A long opening explains too much before the caller can state a need. Keep the greeting brief and let the caller speak.
Asking every caller the same questions
Different intents need different details. Collecting a full lead form for an existing customer wastes time and creates frustration.
Mixing business facts with behavior rules
Hours, services, and locations belong in the knowledge base. Instructions about asking, routing, booking, and escalating belong in the script. Separating them makes updates safer.
Leaving fallback behavior vague
“Get help if needed” is not an operational rule. Define what triggers the fallback, where the call goes, what happens if the handoff fails, and what the receptionist tells the caller.
Promising more than the workflow can deliver
Do not promise a booking, transfer, callback time, or answer unless the system and the team can support it. A clear limitation followed by a useful next step is better than an unsupported promise.
Skipping regression tests after a change
A small edit can affect multiple call paths. Retest the changed path and the related paths before treating the new version as ready.
Frequently asked questions
How long should an AI receptionist script be?
It should be long enough to define the important decisions and short enough for an operator to review. Organize it into role, intents, required details, actions, boundaries, fallback behavior, and confirmation rules. If it becomes difficult to maintain, split it into related call flows instead of adding more general instructions.
Can I copy an AI receptionist script word for word?
You can copy the structure, but the business rules must be specific to your operation. Replace sample services, questions, destinations, schedules, knowledge, and expectations with approved information. Then test the result with the calls your business actually receives.
What should happen when the AI receptionist cannot answer?
The script should move to a defined fallback. Depending on the business rule, that may be a transfer, detailed message, callback request, or another approved route. The receptionist should not guess, and it should tell the caller what will happen next.
How often should I update the script?
Update it when hours, services, staff responsibilities, booking rules, transfer destinations, or common caller needs change. Also review failed calls and recurring fallback events. Retest the related paths after every material change.
Turn the template into a working call flow
A useful AI receptionist script starts with the calls that matter most, gives each one a clear outcome, and defines what happens when the normal path breaks. Build the first version around real caller needs, test it, and refine it from evidence.
If you want to see how these rules work with routing, booking, knowledge, and fallback paths, book a Telvana demo. Bring your most common call types and the next step you want for each one.