WhatsApp voice AI agent: let AI answer and make WhatsApp calls

A WhatsApp voice AI agent is an AI that answers the calls customers place to your WhatsApp Business number and, with permission, calls customers back. It listens, understands the request in natural speech, answers from your knowledge and tools, books or changes things in your systems, and hands the call to a person when it should. Everything runs on the WhatsApp Business Calling API, so the call sits next to the chat on the same number.

Customers already tap the call icon in your chat, and every unanswered call is a lost booking. This guide covers what a voice agent can do, when voice is better than text, the call permission rules Meta sets, how to write the script, the guardrails and handover rules to set, and how to measure the result.

What does a WhatsApp voice AI agent do?

Meta’s calling documentation is direct about this: Meta delivers the raw audio stream of the call, and your business decides how to process it, including with AI voicebots. Meta also points to the WhatsApp Business Solution Terms for restrictions on AI use, which we cover in our WhatsApp AI chatbot policy guide.

In practice, a well-built voice agent does four jobs:

1. Answers. It picks up right away, including at peak times and after hours, greets the caller and asks what they need.
2. Qualifies. It asks the questions your team would ask: what product, which location, budget, timing, order number.
3. Acts. It checks availability, books or moves an appointment, looks up an order, or creates a lead or ticket through connected tools.
4. Hands over. When a call needs a person, it transfers to the right team and passes a short summary so nobody asks the caller to repeat themselves.

A short example of a booking call:

Agent:  Hi, this is the virtual assistant for Lumen Dental. I can book, move or cancel appointments, or connect you to the team. How can I help?
Caller: I need a cleaning next week.
Agent:  I have Tuesday at 10:00 or Thursday at 16:30. Which works better?
Caller: Thursday.
Agent:  You're booked for Thursday at 16:30. I'll send the details in this chat now. Anything else?

When does voice beat text?

WhatsApp chat already handles most service requests well. A voice agent earns its place in specific situations:

Situation	Better on voice	Better on text
Caller is driving, cooking or walking	Yes
Urgent or emotional issue (lockout, damaged delivery, missed flight)	Yes
Long explanation that is hard to type	Yes
Older customers or low literacy	Yes
Sharing order numbers, addresses or links		Yes
Comparing options with photos or a catalog		Yes
Confirmations the customer needs to keep		Yes

The best setups use both. Meta’s documentation notes that you can keep sending messages while a call is in progress, so the agent can talk the caller through a choice and drop the confirmation, the address or the payment link into the chat at the same moment. If you are deciding between bots and agents in general, our comparison of an AI agent vs a chatbot explains the difference.

Inbound calls: what to set up first

Customer-initiated calls are free under Meta’s calling pricing, and they need no permission. Before a voice agent takes them, check these settings from Meta’s call settings documentation:

• Call icon visibility. You decide whether the call button shows in the chat and on your profile, and in which countries.
• Business calling hours. Outside hours, WhatsApp can steer people to message you instead of calling. With an AI agent answering, you may extend hours, but keep a plan for calls that need a person when your team is offline.
• Callback requests. Customers can ask for a callback when you do not pick up or are closed.
• Voicemail. Meta’s settings can answer missed or rejected calls, play your announcement and deliver the voicemail to you as an audio message.
• Call buttons and deep links. Add a call button to messages, or a deep link on your website, with a context string so the agent knows whether the caller came from your pricing page or a delivery notice.

Pick-up rate matters. Meta’s documentation says that if many calls to your number go unanswered, you are warned and calling can be restricted, with the call button hidden. High negative feedback, such as blocks and reports, can also restrict calling.

Outbound calls and call permission

Business-initiated calls are where most of the rules sit. Your business can only call a WhatsApp user who has granted call permission, and only the user can grant or revoke it. According to Meta’s user call permissions documentation:

• You can ask for permission with a free-form interactive message inside an open customer service window, or with an approved template outside it. Templates must include body text that explains why you are asking.
• A customer who calls you grants temporary permission automatically if you turn on the callback setting, and customers can also grant permission from your business profile.
• Permission is temporary (7 days) or permanent, and the customer can change it at any time.
• You can send at most 1 permission request in 24 hours and 2 in 7 days to the same person. The limits reset when a call connects.
• After 2 consecutive unanswered calls the customer is asked whether to keep your permission; after 4, it is revoked automatically.
• Each business number can make up to 100 connected calls every 24 hours to the same user.

Business-initiated calling is not available for business numbers in the United States, Canada, Egypt, Vietnam and Nigeria. Outbound calls are charged per six-second pulse by destination country, and permission requests are billed as normal messages. Our WhatsApp Business Calling API guide covers setup and pricing in more detail.

For a voice agent, this means outbound calls should follow a moment the customer will recognize:

Hi Elena, your car is ready earlier than planned. Our assistant can call you to pick a collection time, it takes about a minute. Is that OK?

[Allow calls]  [Not now]

Good outbound jobs: confirmations, reminders the customer opted into, requested callbacks, and qualifying leads who asked to be contacted. Cold calling people who never asked is the fastest way to collect blocks.

One conversation, one customer record

A call that lives in a separate phone system creates two versions of the customer. On WhatsApp, calls and chats share the same number, and Meta’s pricing documentation says a call from the customer, even an unanswered one, opens or refreshes the 24-hour customer service window, as does a customer accepting your call. That makes it natural to treat them as one conversation:

• The agent sees the customer’s recent messages before answering, and can greet them by name.
• After the call, a summary and outcome attach to the same conversation and contact.
• If the customer writes later, whoever replies sees what was agreed on the call.
• Your CRM gets one update with the result, not a phone note in one tool and a chat log in another.

How to design a voice agent script

Voice is less forgiving than chat. People cannot scroll back, and long sentences get lost. Write the agent’s instructions with these rules:

1. Open with who and what. Name the business, say it is a virtual assistant, list two or three things it can do, and say how to reach a person.
2. One question at a time. “What day suits you?” then “Morning or afternoon?” beats a three-part question.
3. Offer two options, not five. Read out two slots or two plans; send the full list in the chat if needed.
4. Confirm before acting. Repeat names, dates and amounts back: “Thursday the 23rd at 16:30, correct?”
5. Keep answers short. One or two sentences, then a question or a next step.
6. Spell out what happens next. “You’ll get a confirmation in this chat in a moment.”
7. Plan for silence and noise. Re-ask once, then offer to continue in the chat or transfer.

Base every answer on your own content. The same knowledge base that serves your chat agent should serve voice; our guide to building an AI agent knowledge base applies here unchanged.

Guardrails and handover rules

Decide before launch what the agent must never do and when it must pass the call on.

Hand over immediately when:

• the caller asks for a person, in any wording
• the caller is angry or upset after one attempt to help
• the topic is a complaint, a refund outside policy, a legal question or a health question
• the agent fails to understand twice in a row
• the caller is a VIP or has an open escalated case

Never let the agent:

• take card numbers or security codes by voice; send a secure payment link in the chat instead
• promise discounts, refunds or delivery dates your systems have not confirmed
• give medical, legal or financial advice
• invent an answer it cannot find in your knowledge or tools

Pass this to the person who takes over: caller name, reason for the call, what the agent already checked or did, and what the caller is waiting for. The AI-to-human handoff guide has message templates and routing patterns you can reuse for voice.

If you record or transcribe calls, the rules on notice and consent apply too. We cover them in WhatsApp call recording and transcription.

How to measure a WhatsApp voice AI agent

Track a small set of numbers weekly and review a sample of real calls:

Metric	What it tells you
Answer rate	Share of incoming calls picked up; protects you from pick-up restrictions
Resolved by AI	Calls completed without a person, checked against callbacks and repeat contacts
Transfer rate and reasons	Where the agent’s knowledge or tools fall short
Time to a person	How long callers wait after asking for a human
Permission acceptance rate	Whether your outbound requests are timely and clear
Outbound pickup and blocks	Whether customers welcome your calls
Bookings, leads or orders per call	The business result
CSAT after the call	How the experience felt

Meta’s WhatsApp Business account API also returns call analytics such as cost, completed call counts and average call duration, which helps you reconcile what you pay for outbound calls. Before launch, test the agent the way you would test a chat agent, with real questions, edge cases and transfer tests; see how to test an AI agent before launch.

How to run a WhatsApp voice AI agent in MonoChat

In MonoChat, WhatsApp calls are part of the same inbox as your chats, and an AI voice agent can answer them.

• Connect and turn on calling. You connect your WhatsApp Business API number and turn on calling where it is available for your account.
• Route calls to AI or agents. You decide which calls the AI voice agent answers and which go to a team. When customers call, MonoChat can identify the customer, understand the topic and route the call with the conversation history attached.
• Let AI handle the routine. The voice agent answers repetitive questions, collects missing information, explains next steps and checks your knowledge and tools, then hands over to a person when the call needs personal attention.
• Trigger outbound calls from workflows. Workflow automation can start WhatsApp calls for confirmations, reminders, follow-ups or lead qualification from events, forms and CRM changes, and can ask for call permission in the chat before an outbound call.
• Summarize and sync. After each call, MonoChat can summarize the outcome, tag intent and sentiment, and push notes to your CRM or helpdesk.

The same AI voice works on your regular phone line. AI voice calls over SIP connect your SIP trunk for inbound and outbound calls with IVR, queues and routing rules, and an AI receptionist answers, books and transfers calls with a transcript and summary saved in the inbox. You write its greeting and instructions, connect your calendar, CRM or helpdesk through AI function tools and workflows, then review transcripts and refine. Larger teams add skill-based routing and SLA dashboards with the multi-agent call center.

For AI costs, see the AI cost calculator and the pricing page. You can start on the free Basic plan (one agent, one channel) at app.monochat.ai, or talk to us about calling for your number.

Sources

• Meta for Developers: WhatsApp Business Calling API overview
• Meta for Developers: User call permissions
• Meta for Developers: Business-initiated calls
• Meta for Developers: User-initiated calls
• Meta for Developers: Configure call settings
• Meta for Developers: Calling API pricing
• Meta for Developers: Session Initiation Protocol (SIP)
• Meta for Developers: Calling FAQ
• WhatsApp Business Solution Terms