
- Meta's terms ban general-purpose AI assistants whose main product is the AI; a business assistant for your own customers is a different use.
- Free-form replies are allowed only inside the 24-hour customer service window; after it closes, only approved templates can be sent.
- You need opt-in before starting any conversation, and every request to stop must be honoured.
- The Messaging Policy requires a prompt, clear and direct way to reach a person when you use automation.
- WhatsApp Business Platform data may not be used to train AI models, except fine-tuning for your business's exclusive use.
Can a business run an AI assistant on WhatsApp?
Yes, when the assistant serves your own customers as part of your business: answering questions, checking orders, booking appointments and passing conversations to your team. What Meta has banned is general-purpose AI assistants whose main product is the AI itself. Your assistant still has to follow the WhatsApp rules for every business: opt-in before you message people, free-form replies only inside the 24-hour customer service window, approved templates outside it, and a clear way to reach a person.
This guide covers the WhatsApp Business Platform (the Cloud API), not the WhatsApp Business app on a single phone. It is general information, not legal advice, and Meta applies its own terms at its discretion.
What exactly did Meta ban?
The rule sits in section 4.7, "AI Providers", of the Meta Terms for WhatsApp Business Platform, formerly the WhatsApp Business Solution Terms. Meta's AI providers page dates the change to January 15, 2026.
- Who is covered: developers of AI and machine learning technologies, such as large language models and generative AI platforms, as Meta determines.
- What is banned: those providers may not use the platform, directly or indirectly, when the AI is the primary functionality being offered. Meta decides what counts as primary.
- Exceptions: the AI providers page lists markets where Meta says it is legally required to allow general-purpose assistants, and even there they may only reply inside an open 24-hour window.
Two parts of section 4.7 matter for an ordinary business. The terms say a business may still retain an AI provider as its solution provider, so using an AI model to power your own assistant is a recognised setup. They also say WhatsApp Business Platform data may not be used to create, develop, train or improve AI models, with one exception: fine-tuning a model for the business's exclusive use.
How does the 24-hour window shape the assistant?
Meta's messaging docs explain that when a customer messages or calls you, a 24-hour customer service window starts, and it resets each time they message again. While it is open, you can send free-form service messages. When it closes, you can only send pre-approved template messages.
For an AI assistant, that means three design rules:
- Replies happen inside the window. The assistant answers while the customer is talking to you, which is where most of its value is.
- Follow-ups use templates. A booking reminder or a delivery update sent a day later must be an approved template. The templates overview lists three categories: authentication, marketing and utility, and a template must be approved before it can be sent.
- Your server checks the window before sending. If the window has closed, the assistant should switch to the right template or hand the case to a person, never send a free-form message that will fail.
What opt-in do you need?
The WhatsApp Business Messaging Policy says you may only contact people on WhatsApp if they have given you their number and you have received opt-in permission confirming they wish to receive messages from you. A customer who writes to you first has started the conversation, but any message you start later, such as a reminder or an offer, needs that opt-in.
Collect opt-in where people already share their number: a checkout field, a booking form or a signup page, with plain wording that names your business and the kind of messages you will send. Store when and where each person agreed, and honour every request to stop. Our guide to the rules lead-generation AI agents must follow covers consent in more depth.
How should handoff to a person work?
This is a written rule, not just good practice. The Messaging Policy allows automation when responding during the 24-hour window, but requires prompt, clear and direct escalation paths, such as an in-chat transfer to a human agent, a phone number, email, web support, an in-store visit or a support form.
- Let the customer ask for a person at any point, in their own words.
- Hand over automatically when the assistant is unsure, the customer is upset, or the topic involves money, complaints or personal data.
- Pass the conversation and a short summary into your team inbox or CRM, so nobody asks the customer to repeat themselves.
- Outside working hours, say when a person will reply instead of leaving the customer waiting.
Test these paths before launch, as we describe in testing an AI assistant before launch.
What does a WhatsApp AI assistant cost to run?
There are three cost lines. The first is Meta. Its pricing page says Meta has charged per message since July 1, 2025. From October 1, 2026, each business phone number gets a free tier of 1,000 service messages per month, and service messages after that are charged. Template messages are charged by category. Check the page on the day you budget, because the rules have changed several times.
The second is the AI model, billed by the tokens each conversation uses. The third is hosting and, if you use one, a solution partner's fees. Ask any vendor to show each line separately.
How do you build one?
You can connect to the Cloud API directly or through a solution partner. Meta's solution provider overview describes partners as Meta Business Partners offering a full range of platform services, while a business building only for itself can follow the Cloud API guide. A partner adds a ready-made inbox and tools; a direct build gives you full control of the data and the logic.
A typical build looks like this:
- Webhook: your server receives each incoming message and records the window start time and opt-in status.
- Knowledge: the assistant answers from your own approved content, such as policies, services and FAQs, using one of the methods in our guide to giving an AI model your business knowledge.
- Actions: order lookups, bookings and lead capture go through your systems' APIs, with limits on what the assistant may change.
- Handoff: conversations move into the team inbox or CRM, with the summary attached.
- Logs: every reply, template and handoff is recorded for review.
Our AI assistants and chatbots team builds the assistant and its knowledge, our workflow automation team connects bookings and orders through n8n, Make or custom code, and our CRM and lead systems team routes handoffs and leads into HubSpot or GoHighLevel.
Frequently asked questions
Did Meta ban chatbots on WhatsApp?
No. Since January 15, 2026, Meta's terms ban general-purpose AI assistants offered by AI providers when the AI is the main product. A business assistant that answers questions about your own services, orders and bookings is a different use. Meta decides how its terms apply, so keep the assistant focused on your business.
Can the assistant message customers first?
Only with opt-in and only with an approved template message. Free-form replies are allowed inside the 24-hour window that opens when the customer messages or calls you. Reminders, updates and offers sent later must use templates in the right category, and marketing messages need the customer's agreement.
Do we need a solution partner, or can we connect directly?
Either works. A business can build on the Cloud API itself, or use a Meta solution partner that adds hosting, an inbox and tools. Choose based on who should own the data and logic, how many agents need an inbox, and whether you already use a CRM that connects to WhatsApp.
Does the assistant have to say it is an AI?
We found no WhatsApp rule that requires a bot to announce itself, but the policy does require a clear route to a person, and other laws may require disclosure depending on where your customers are. Telling people they are talking to an assistant, and how to reach a person, is the safe default.
Can we use WhatsApp conversations to train our AI model?
Section 4.7 of Meta's terms says WhatsApp Business Platform data may not be used to create, develop, train or improve AI models, except to fine-tune a model for your business's exclusive use. Data protection law also applies, so get advice before reusing customer conversations for any training.
Last updated