AI on WhatsApp: what Meta allows, and what it does not
Editorial Team
5 min read
Meta’s WhatsApp Business Solution Terms now prohibit AI providers from using the platform to distribute general-purpose assistants. A ChatGPT-style bot that answers anything cannot run on the Business Platform. A bot scoped to your own business — order status, bookings, product questions, support — remains permitted.
The distinction is not about whether you use AI. It is about whether the AI is the product or the plumbing.
What exactly do the terms say?
Meta added a section addressing “AI Providers” to its Business Solution Terms. It prohibits providers and developers of AI or machine learning technologies — large language models, generative AI platforms, general-purpose AI assistants — from using the WhatsApp Business Solution to make such technologies available where they are the primary, rather than incidental or ancillary, functionality being offered.
Meta determines what falls inside that definition at its own discretion, which is worth noting: the boundary is not one you can argue your way across.
Its stated reasoning is that the Business Platform was built for companies to communicate with their customers, not to serve as a distribution channel for standalone AI assistants. Several well-known assistants were withdrawn from WhatsApp as a result, and the change has since attracted antitrust scrutiny from regulators in Europe and Brazil.
You can read the clause yourself in Meta’s WhatsApp Business Solution Terms, and should, before making a decision on it.
Two exceptions worth knowing about
You may still retain an AI provider as your service provider. The terms permit a business to engage an AI provider as its Third Party Service Provider. The prohibition is on the provider distributing assistants through WhatsApp as their own product — not on you using their technology to run your own customer service.
That distinction is the one most businesses actually need, because almost nobody builds their own model. Using someone else’s to power a scoped business bot is a different thing from that provider offering an assistant on WhatsApp.
There is a regional carve-out. The clause allows such technologies to be made available to WhatsApp users with European Economic Area or Brazil country codes. If your customers are in India, this does not help you — but it explains why coverage of the change differs depending on where it was written.
Where is the line?
| Not permitted | Permitted |
|---|---|
| An open-domain assistant that answers any question | A bot answering questions about your products and services |
| AI capability sold as the service itself | AI supporting a defined business process |
| A general assistant distributed to consumers via WhatsApp | Order tracking, bookings, support, notifications |
| “Ask me anything” as the core proposition | Structured workflows with AI handling the language |
A practical test: if a customer asked your bot for a recipe, or for help with their homework, what should happen? A compliant business bot says that is outside what it can help with. A general-purpose assistant answers. If yours would answer, that is the problem.
Does this affect most businesses?
Directly, no. The clause targets AI providers distributing assistants, not ordinary businesses automating their own customer conversations.
Indirectly, it matters more than it appears. Plenty of businesses have built support bots by connecting a language model to WhatsApp with a loose instruction and no boundaries. Those bots answer whatever they are asked, because nobody constrained them. That is a general-purpose assistant in everything but intention.
Worth checking whether any of these are true of your setup:
- The bot will attempt any question rather than declining out-of-scope ones
- There is no defined scope written down anywhere
- It was built by pointing a model at WhatsApp with a short prompt
- Customers use it for things unrelated to your business, and you find that amusing rather than concerning
Scoping such a bot properly is not a large project. It is a boundary instruction, a refusal behaviour and a handover route — the same three things that make a bot useful anyway.
What about the cost of running one?
This changed on 1 October 2026, and it affects bots more than people.
Until then, non-template messages sent inside an open 24-hour customer service window were free, as were utility templates delivered within one. From 1 October 2026 both are charged, and service became a fourth billable message category alongside marketing, utility and authentication. Meta confirms this in its pricing explainer, published alongside the platform pricing documentation.
An automated agent is affected more than a human team, because bots send more messages — acknowledgements, clarifying questions, follow-ups. Every one of those is now a charge, on every conversation, all day.
A bot that resolves a query in six messages where a person would have taken two costs three times as much. The metric to watch is messages per resolved conversation, not conversations handled — and most bot dashboards do not report it.
That was worth measuring before the change, because a chatty bot is a worse experience. It is now worth measuring twice over. We have covered the billing change in full here.
What should a WhatsApp bot actually do?
Fewer things, well.
Answer within a stated scope. Your products, your policies, your orders. Decline everything else politely and hand over.
Reach conclusions in few messages. Ask for everything needed in one message rather than three sequential questions. Better service, and cheaper if the billing changes.
Hand over cleanly. With the conversation attached, and quickly — a customer who has asked twice for a person should get one.
Say when it does not know. The behaviour that keeps you compliant and the behaviour that keeps customers trusting it happen to be the same behaviour.
Common questions
Can we still use a language model on WhatsApp at all?
Yes, provided it serves a defined business function rather than being offered as a general assistant. Support, sales, bookings and order tracking are the sort of use the platform exists for, and Meta has said explicitly that businesses using AI to serve their own customers are not the target.
We use a third-party AI platform. Is that a problem?
Not in itself. The terms permit engaging an AI provider as your Third Party Service Provider. What is prohibited is that provider distributing its own assistant through WhatsApp as the primary product.
What happens if we are non-compliant?
Enforcement is at Meta’s discretion and can extend to restriction of the account. Given that your WhatsApp number is often a primary customer channel, that is not a risk worth carrying for the sake of an unscoped bot.
Does this apply to the WhatsApp Business App?
The terms concern the Business Solution — the API. The free app has always had limited automation and is not what the change addresses.
Should we move our bot to another channel?
Only if your use genuinely is a general assistant. For business support, WhatsApp remains where Indian customers actually are, and a properly scoped bot is permitted. Scoping it is cheaper than migrating your customers.
Platform terms change, and parts of this area are disputed. Read Meta’s current Business Solution Terms before relying on any summary, including this one.
More on this: building a chatbot that answers from your documents and should your business use AI for support. See also WhatsApp Business API.
Keep reading
More on AI & Automation
What everyone gets wrong about MCP and APIs
The two are not alternatives. Understanding what sits where decides whether you are adding a useful layer or a new attack surface.
You did not just upload a selfie
The 80s photo trend is fun. The version taken at your desk, with a client name on the screen behind you, is a different thing.
When a language model is the wrong tool
A great deal of what gets proposed as AI work is better served by a database query, and saying so is the fastest way to tell whether your vendor is worth listening to.