← Guides

        Guide
        
        WhatsApp

How to automate WhatsApp Business with AI: three routes and when each one fits

In 2026 there are three ways to automate WhatsApp Business with AI. The app's own tools and Meta's AI agent, if your questions are few and simple. A bot you set up yourself on the API, if you have frequent questions and someone on the team to maintain it. And an agent built on your rules and connected to your systems, if every message has to end in a case a person decides on.

Thomas Trincado Updated 30 September 2026

Topic WhatsApp

Read 9 min

Before you start: what do you want to automate?

"Automating WhatsApp" means different things to different people. It is almost always one of these three jobs, and the job decides the route more than the tool does:

Answering questions that repeat: opening hours, prices, requirements, lead times, "how is mine going?".
Collecting data and documents: asking for the registration number, the ID or the missing photos before a person looks at the case.
Booking or opening a request: the conversation ends in an appointment, an order or a case inside your system.

Count, too, how many messages you get a day and who answers them now. With twenty messages and one person, route 1 is usually more than enough. With hundreds and a team, it is not.

If what you want to automate does not start on WhatsApp, such as email across several inboxes or checking case files, see AI process automation and AI document review.

Route 1: what the WhatsApp Business app already does

The free WhatsApp Business app already has simple automation: a business profile, a catalogue, greeting and away messages, quick replies and labels to sort chats (app page and Meta documentation, read 30 Sep 2026). You can use it on one phone and up to four linked devices (WhatsApp Help Center).

What is new is Meta's AI. On 3 June 2026 Meta announced it was opening Meta Business Agent to businesses of all sizes worldwide: it answers questions about your business, recommends products from your catalogue, books appointments and qualifies leads, and you decide when a person steps in. Meta says getting started is free and that access will move to paid subscriptions.

It fits when your questions are general and the answer is in public information: your catalogue, your prices, your hours. It falls short when the answer depends on your own rules, when documents need checking, or when the conversation has to open something in your management software. We go into detail in Meta Business Agent or a custom assistant.

Route 2: a bot you set up yourself

The second route is to move your number to the WhatsApp Business API and set up a bot yourself. There are two versions:

A chatbot platform on the API. This is what Meta suggests if you have no developers: a provider that connects WhatsApp to an omnichannel inbox or a bot platform (WhatsApp Business FAQ). You build the flows in their editor, without code.
Do it yourself with n8n and an AI model. Automation tools such as n8n have a node for the WhatsApp Cloud API that sends messages and templates and uploads or downloads media (n8n documentation). You connect it to a model such as Claude or GPT and write the instructions.

It works well for a team with time to maintain it: writing and testing the flows, reading conversations, fixing wrong answers, and keeping to Meta's rules, which require a clear way to reach a person. What does not show on the invoice is that time.

What is the best chatbot for WhatsApp Business? There is no single one. The best is the one your team will actually maintain, with the features your questions need. If nobody on your team will review it every week, this route degrades fast.

Route 3: a custom AI agent

The third route is an AI agent built around your process, not a configured template. What sets it apart:

It answers from your rules, criteria and exceptions, written and reviewed with you.
It reads photos, documents and voice notes, and checks them against your list of requirements.
It opens the request in your own system, through its API, with the data your team needs.
It hands whatever needs judgement to a person, with the whole conversation attached, and picks up again when they are done.
It is tested against a set of real cases before launch.

That is how the assistant we built for Fixlab, a vehicle certification lab, works. Between 26 June and 24 September 2026 it handled 279 conversations from 218 people, with a median first reply of 16 seconds. 55% started outside office hours, 51% closed without a person, and it opened 20 reform requests end to end in Fixlab's platform.

To see how it would answer your customers, try the demo with your own messages. What it does and how we build it is on the page for our custom AI agent for WhatsApp.

Table: which route fits you

When in doubt, compare your case row by row. If most of your answers land in one column, that is your route.

	Route 1: the app	Route 2: your own bot	Route 3: custom
Volume	Low: one or two people handle it.	Medium: questions that repeat a lot.	High, or every message is a case.
Kind of question	Hours, prices, availability.	Frequent questions with a fixed answer.	Depends on your rules and exceptions.
Documents and photos	A person looks at them.	The bot collects them; a person checks them.	It reads them and checks them against your requirements.
Connection to your system	None, or the systems Meta connects.	Whatever your team builds with the tool.	Built on your system, including internal software.
Who maintains it	You, in the app.	Your team.	Whoever builds it, with documentation so your team can change answers and rules.
What you pay	The app is free; Meta's agent is free to start, then by subscription.	The platform or your hours, Meta's messages and the AI model.	Build and upkeep, the AI model and Meta's messages.

Odysi comparison based on the sources in this guide. Meta's prices are on its pricing page.

If your case is route 3, tell us how many messages you get and we will tell you honestly whether it pays.

How to put AI on WhatsApp Business, step by step

These are the steps for route 3. If you take route 2 they still apply: they are what separates a bot that works from one that embarrasses you.

Step 1Write down the rules and the answers. What it may answer, what it must never say and when it hands the case to a person.
Step 2Collect real conversations. Build a test set from them that covers every path the conversation can take. At Fixlab it was 50 scenarios.
Step 3Move the number to the API. What you need is in what is the WhatsApp Business API, including the option to keep using the app on the same number.
Step 4Build. Knowledge base, connection to your system and handover to a person with the full conversation.
Step 5Test before launch. Agree the pass mark in writing. At Fixlab it was zero critical failures and at least an 80% pass rate.
Step 6Run a supervised pilot. A few weeks with someone on your team reading the conversations and fixing what goes wrong.
Step 7Launch and measure. Time to first reply, conversations closed without a person and the reason for each handover.

What you should not automate

Automating does not mean answering everything. Always leave these to a person:

Final decisions: approving, rejecting, making a price exception.
Complaints and angry customers.
Anything with legal or financial effect for the customer.
Cases that depend on reading a doubtful photo or fall outside the pattern.

A good assistant is not the one that answers everything, but the one that can tell the difference and hands over instead of guessing. At Fixlab, most handovers to a person are price questions, which the assistant passes to the team by design.

Meta's rules and GDPR

Three Meta rules that apply to every route:

Customer service assistants are allowed; since 15 January 2026 Meta bars general-purpose AI chatbots. We explain it in Meta's rules on AI chatbots.
Any automation needs "prompt, clear, and direct escalation paths" to a person (WhatsApp Business policy).
You may only message first people who gave you their number and agreed to hear from you.

On data protection, Meta's policy is clear that your business is responsible for the notices, permissions and consents needed to collect and use your customers' data. Before launch, document where each component (WhatsApp, the AI model, your server) processes and stores data, so you can show it to whoever handles data protection. In our projects we choose those components with the client and can run the model on an EU endpoint.

What it costs

On routes 2 and 3 you pay Meta per message. From 1 October 2026 that includes service replies, after 1,000 free a month per number (Meta pricing). Rates and a worked example are in what WhatsApp charges per message. An assistant that resolves in a few messages costs less than one that goes round in circles.

On top of that come building and maintaining it, and the AI model's usage. What moves the price most is not the model but what the assistant can do and whether it works inside your system. We explain it in how much a WhatsApp chatbot costs.

Sources

Read on 30 September 2026. Meta changes its products and prices; check each point at the source before you decide. This guide is not legal advice.

FAQ

FAQ: automating WhatsApp Business with AI

How do you automate WhatsApp Business with AI?

There are three routes. The app's tools and Meta's AI agent, for simple questions. A bot you set up yourself on the API, if someone on your team will maintain it. And a custom agent, if the answer depends on your rules or the conversation has to open something in your system.

How do I put an AI assistant on WhatsApp?

If Meta's AI is enough, you turn it on from Meta's business tools. For your own assistant, with your rules or connected to your system, your number has to move to the WhatsApp Business API and the assistant connects to it.

Can you automate WhatsApp Business for free?

With the app, yes: its automatic messages are free, and Meta says its AI agent is free to start. Anything on the API pays Meta per message (with 1,000 free service replies a month per number), plus the tool or the build.

Can I automate WhatsApp with n8n or Claude?

Yes. n8n has a node for the WhatsApp Cloud API and you can connect it to a model such as Claude. You need the number on the API and someone to build, test and maintain the flows.

What is the best chatbot for WhatsApp Business?

The one your team will maintain and that covers what your customers ask. For general questions, Meta's; if you need flows and have someone to look after them, a platform; if the answer depends on your rules or your system, a custom one.

Does WhatsApp allow AI chatbots?

Yes, the ones that serve a business's own customers. Since 15 January 2026 Meta bars general-purpose AI chatbots. We explain it in Is AI allowed on WhatsApp Business?