How to build a WhatsApp chatbot (or AI agent) that actually helps customers

A WhatsApp chatbot is software that answers customers on your WhatsApp Business number automatically. Simple bots follow rules and menus; AI chatbots understand free-text questions and answer from your own content. To build one that actually helps, connect your number to the WhatsApp Business API, start from the five or ten questions customers ask most, ground the answers in your real policies and data, and give every conversation a clear way to reach a person.

This guide covers the types of WhatsApp bots, a step-by-step build plan, the Meta rules that apply in 2026, what it costs to run, and a sample flow you can copy.

What can a WhatsApp chatbot do for a business?

A good bot takes the repetitive work off your team and gives customers an answer at any hour. Typical jobs:

• Answer frequently asked questions about prices, hours, delivery, returns and policies
• Check order status or booking details from your systems
• Book, move or cancel appointments
• Qualify leads by asking a few questions before a salesperson joins
• Collect structured details such as an address, a photo of a damaged item or a form
• Route the conversation to the right team when a person is needed

What a bot should not do is trap people. If customers have to fight the bot to reach someone, the chatbot is costing you goodwill, not saving time.

Rule-based vs AI chatbots: which one do you need?

	Rule-based chatbot	AI chatbot	Hybrid
How it works	Keywords, buttons and list menus lead to fixed answers	A language model reads the message and answers from your content	Menus for common paths, AI for everything else
Handles free text	Poorly	Well	Well
Setup effort	Map every path	Add content, write instructions, test	Both, but each part is smaller
Predictability	Very high	Needs grounding and testing	High on key paths
Best for	Narrow tasks: order status, booking, opt-outs	Many topics, many languages, long-tail questions	Most businesses

On WhatsApp, rule-based bots lean on interactive messages. Meta’s Cloud API supports reply buttons (up to three per message) and list messages (up to ten rows across all sections). These keep answers structured, which makes them easy to use in automated flows. AI chatbots use a large language model plus a knowledge base, so they can understand “my parcel never came, what now?” without a keyword match.

Most teams end up with a hybrid: buttons for the three or four journeys that matter most, and AI for the long tail of questions nobody thought to put in a menu.

How to build a WhatsApp chatbot, step by step

1. Get API access for your number

Chatbots run on the WhatsApp Business API (the WhatsApp Business Platform). You connect your own WhatsApp Business account directly to Meta’s Cloud API, for example with MonoChat, and if you already use the WhatsApp Business app, Coexistence lets you keep the app on the same number while the bot works through the API.

2. Pick the jobs, not the technology

Export a month of conversations and group them. You will usually find that a handful of topics make up most of the volume. Write those down with the ideal answer for each. This list becomes your test set later.

3. Prepare your content

An AI chatbot is only as good as what it can read. Collect your return policy, delivery terms, price lists, opening hours, product details and FAQs. Remove outdated versions. If a document contradicts another one, the bot will too.

4. Write clear instructions

Tell the assistant who it is, what it may help with, what it must not do, and when to hand over. For example:

You are the WhatsApp assistant for [Business].
Answer only questions about our products, orders, delivery and returns.
Use only the knowledge provided. If the answer is not there, say so and offer a person.
Never ask for full card numbers or ID numbers.
Hand over to the support team when the customer asks for a person,
is upset, or mentions a refund above [amount].
Reply in the customer's language. Keep answers under 80 words.

5. Connect your systems

Answers like “where is my order?” need live data. Connect the bot to your store, CRM or booking system through integrations or function calling, so it checks before it promises.

6. Design the handover

Decide exactly when a person takes over and what they see. A good handover passes the full chat, the customer’s details and the reason for escalation, and tells the customer what happens next.

7. Test with real questions, then launch small

Run your list from step 2 through the bot. Check every answer against your policies, try awkward phrasing and other languages, and test what happens when the bot does not know. Launch on a share of traffic or a single topic first, read the conversations daily for the first weeks, and widen the scope as accuracy holds.

A sample WhatsApp chatbot conversation

Here is a hybrid flow for an online store. The customer types freely; the bot uses buttons where a choice is needed.

Customer: hi my order hasn’t arrived yet Bot: Sorry about the wait. Could you send your order number? It starts with #. Customer: #48213 Bot: Thanks. Order #48213 left our warehouse on Monday and the carrier shows it out for delivery today. Would you like the tracking link or help with something else? [Send tracking link] [Change delivery] [Talk to a person] Customer: Talk to a person Bot: Of course. I have passed your order details to our support team. Someone will reply here shortly, usually within 15 minutes during opening hours.

Notice what the bot does well: it asks for one thing at a time, checks real data, offers the next step as buttons, and hands over without friction.

What Meta’s rules mean for your chatbot in 2026

The 24-hour window and templates

Your bot can send free-form replies only within the 24-hour customer service window that opens with each customer message. To start a conversation, or to follow up after the window closes, it needs an approved template message, and you may only contact people who opted in.

Escalation is mandatory

The WhatsApp Business Messaging Policy allows automation inside the window but requires prompt, clear and direct escalation paths: an in-chat transfer to a human agent, a phone number, email, web support, an in-store visit or a support form.

Business AI is fine, general-purpose AI assistants are not

Since January 15, 2026, WhatsApp’s Business Solution Terms bar “AI Providers” from using the platform when the AI itself is the primary product, such as a general-purpose assistant, except for users in markets where Meta is legally required to allow it. AI that supports your own business, like answering questions about your products or booking appointments, is incidental to your service and remains allowed. The terms also restrict using WhatsApp business data to train AI models, with a narrow exception for fine-tuning a model for your own exclusive use. Our WhatsApp AI chatbot policy 2026 guide explains the details.

Do not collect sensitive data in chat

The policy says businesses must not ask people to share full payment card numbers, financial account numbers, ID card numbers or other sensitive identifiers. Use a secure payment link or form instead.

How much does a WhatsApp chatbot cost?

You pay for three things:

1. The platform that runs the bot and the inbox. See pricing for MonoChat plans.
2. The AI model, if the bot uses one. With your own API key you pay your model provider directly for usage.
3. Meta’s message fees. Meta charges per delivered message by category and recipient country. Since October 1, 2026, service messages (free-form replies inside the window) are charged after the first 1,000 per business phone number each month. Messages from customers to you are never charged.

Example: a chatbot that sends 6,000 replies a month to customers in Brazil from one number pays Meta for 5,000 of them. At the October 2026 service rate of $0.0068, that is about $34. The same volume to Germany at $0.055 is about $275. Because each message bubble counts, bots that answer in one well-written message cost less than bots that send three short ones. Try your own numbers in the bulk messaging calculator or estimate the model side in the AI cost calculator.

What about Meta’s own AI agent?

Meta now offers Meta Business Agent, an AI agent configured on its own platform. It is a quick start for eligible businesses, but Meta’s developer documentation excludes several verticals, including finance, government, health, alcohol and gambling, requires that no other AI agent is active on the number, and its messages are billed per token rather than at the service rate. If you need your own model, your own tools or one inbox across several channels, a platform-based bot gives you more control.

Mistakes that make WhatsApp chatbots fail

• No exit. Customers who cannot reach a person leave, or complain elsewhere.
• Guessing. An AI bot without grounding in your content can invent policies. Restrict it to your knowledge and let it say “I don’t know”.
• Too many bubbles. Splitting one answer into several messages is annoying and, since October 2026, costs more.
• Menus ten levels deep. If a path needs more than three taps, use free text and AI instead.
• Set and forget. Read conversations weekly, add missing answers and retire outdated content.

How to build your WhatsApp chatbot in MonoChat

MonoChat puts an AI chatbot on your WhatsApp Business number and keeps your team one step away in the same inbox. The setup follows four steps:

1. Connect your WhatsApp number to MonoChat through the API. With Coexistence you can keep the WhatsApp Business app on the same number.
2. Add your content to the AI knowledge base: PDFs, Excel files, web pages and text, organized into scopes such as support, sales or policies, so the bot answers from your documents instead of guessing.
3. Set the rules. Choose the AI model (OpenAI, Anthropic, Google Gemini or your own, with your own API keys, see custom LLMs), write the instructions and decide when the bot books, starts a workflow or hands over.
4. Go live and improve. Follow conversations in the unified team inbox and use analytics to refine answers.

Guided paths such as booking, order status or opt-outs are built in the no-code workflow automation builder with buttons, lists, conditions and calls to your systems, and WhatsApp Flows collect structured details in in-chat forms. When the bot cannot help or the customer asks for a person, the conversation is routed to the right agent with the full history, and the agent replies from the same WhatsApp number.

The same assistant can also answer on Instagram, Messenger, web chat and other channels you connect. See the WhatsApp chatbot page for details, or start on the free Basic plan at app.monochat.ai. If you only need simple greeting and after-hours replies for now, start with our guide to WhatsApp auto reply messages.

Sources

• WhatsApp Business Messaging Policy
• WhatsApp Business Solution Terms
• Pricing on the WhatsApp Business Platform (Meta for Developers)
• Interactive reply buttons messages (Meta for Developers)
• Interactive list messages (Meta for Developers)
• Meta Business Agent Platform overview (Meta for Developers)