jabee

Meta banned AI chatbots from WhatsApp. Yours is probably fine — here is how to check

Last updated 28 October 2026

By Mumen Tayyem, founder of Jabee

Short version. Meta's terms for the WhatsApp Business Platform prohibit providers of AI technologies from using the platform to deliver those technologies. That is aimed at ChatGPT, Perplexity and the other general-purpose assistants that lived inside WhatsApp, and they have left. A business running an AI assistant to answer its own customers is a different thing, is explicitly contemplated by the same terms, and is not banned. The condition attached to it is about training data, not about automation, and that is the part worth checking with whoever supplies your assistant.

The headlines in October 2025 said "Meta bans AI chatbots from WhatsApp", and a lot of people running a perfectly ordinary WhatsApp assistant for a shop or a clinic read that and assumed they were about to be switched off. Eleven months later most of them still think so. This is what the terms actually say.

What the clause says

The rule lives in Meta's Terms for the WhatsApp Business Platform, under a section headed AI Service Providers. The operative words are that it is "strictly prohibited" for:

providers and developers of artificial intelligence or machine learning technologies, including but not limited to large language models, or generative AI platforms, or general-purpose AI assistants

to access or use the platform to "provide or deliver or display or sell or make available these technologies in any other way".

Read that twice, because the subject of the sentence is the whole point. The prohibited party is a provider of the AI itself, using WhatsApp as a distribution channel for that AI. It is not a business using an AI to answer questions about its own products.

Reporting at the time put the effective date at 15 January 2026, with new API users registering from 15 October 2025 subject to it immediately. Meta's terms page itself carries no effective date for the clause, which is worth knowing if you are trying to pin down exactly when something changed.

The distinction, in one line each

Banned: a company whose product is an AI assistant, distributing that assistant to the public over WhatsApp. Someone messages the number and talks to a general-purpose AI about anything they like. ChatGPT's WhatsApp number was the clearest example, and it is gone.

Allowed: a business whose product is something else, using an AI assistant to answer its customers' questions about that something else. A bakery whose number answers questions about opening hours and delivery. A clinic that books appointments. A car dealer whose assistant knows which models are in stock. The assistant is scoped to one business's own knowledge, and it exists to serve that business's customers rather than to be used as an AI.

The test is not how clever the assistant is, or whether it uses a large language model — a well-built business assistant almost certainly does. The test is whether the AI is the product being delivered to the public, or the means by which one business answers its own customers.

The carve-out, and the condition on it

The same terms allow a business to retain an AI service provider as a solution provider. That sentence is the whole legal basis for every WhatsApp assistant worth having, including ours, and it comes with a restriction that matters more than people expect:

Platform data must not be used to create, develop, train or improve any machine learning or AI system. There is a narrow exception for fine-tuning a model for the business's own exclusive use.

So the thing Meta cares about is not that an AI answered a message. It is whether your customers' WhatsApp conversations ended up as training data. That is the question to put to your provider, and it is a better question than "are you compliant", because it has a factual answer.

Ours is: no. Conversations and documents are used to answer that business's customers and nothing else, they are not used to train models, and there is no fine-tuning step in the product at all — the assistant answers by searching a business's own content at the moment a question arrives, so there is nothing to train on and nothing to leak between businesses.

What to actually check

Four questions, in the order they matter:

  1. Is the assistant scoped to one business's own content? If it will cheerfully answer questions about anything, it looks like the thing the clause prohibits, whoever built it.
  2. Does the provider train on your conversations? Ask for a plain yes or no, in writing. This is the explicit condition in the terms.
  3. Is it on the official API at all? Plenty of cheap "WhatsApp automation" runs on unofficial libraries driving WhatsApp Web. Those were never permitted, have nothing to do with this clause, and are the actual reason numbers get banned. If your provider offered you a QR-code login rather than Meta's own signup window, that is the thing to worry about this week.
  4. Who holds the WhatsApp Business Account? If it is the provider's, your number's standing depends on their compliance as well as yours.

Why this scared people more than it should have

Three reasons, and the third is the interesting one.

The headlines were accurate and incomplete. "Meta bans AI chatbots from WhatsApp" is a fair description of ChatGPT leaving and a terrible description of what it means for a bakery.

The change arrived in the same few months as Meta's own Business Agent and the 1 October service message charge, so three unrelated things became one vague sense that WhatsApp was closing down for business.

And Meta is now a competitor in this market as well as the landlord of it. Reading a rule that removes general-purpose assistants from WhatsApp, in the same year Meta shipped its own assistant for businesses on WhatsApp, it is reasonable to wonder. Worth saying plainly: the clause as written does not privilege Meta's agent over a third-party business assistant, and the carve-out for solution providers is explicit. The thing to watch is not this clause but the eligibility list on Meta's own agent, which excludes whole industries.

Frequently asked questions

Is my WhatsApp AI chatbot banned in 2026? Almost certainly not. The prohibition is on providers of AI technologies using WhatsApp to distribute those technologies, such as ChatGPT or Perplexity. A business using an AI assistant to answer its own customers about its own products is a different arrangement, and Meta's terms explicitly allow a business to retain an AI service provider as a solution provider.

What exactly did Meta ban? Providers and developers of AI or machine learning technologies, including large language models, generative AI platforms and general-purpose AI assistants, from using the WhatsApp Business Platform to provide, deliver, display, sell or otherwise make those technologies available. The subject of the rule is the AI vendor, not the business answering its customers.

When did the WhatsApp AI policy change? Reporting put the effective date at 15 January 2026, with new API users registering on or after 15 October 2025 subject to it from registration. Meta's own terms page does not display an effective date for the clause.

Can I still use ChatGPT on WhatsApp? No. The general-purpose assistants that ran on WhatsApp numbers, including ChatGPT and Perplexity, are the thing this rule removed, and they have left the platform.

Does my provider train AI on my customers' messages? That is the condition attached to the carve-out, so it is the question worth asking in writing. Platform data must not be used to create, develop, train or improve machine learning or AI systems, with a narrow exception for fine-tuning a model for the business's own exclusive use. Jabee does neither: conversations are used to answer that business's customers and nothing else.

Is an AI assistant that books appointments still allowed? Yes. Structured work for one business — answering questions from its own content, tracking an order, booking an appointment, recommending from its own catalogue, capturing a lead and handing over to a person — is the allowed shape. It is also the shape almost every business assistant already has.

What actually gets a WhatsApp number banned? Not this clause. Numbers are restricted for unofficial automation driving WhatsApp Web, for poor quality ratings from people blocking and reporting messages, and for problems in the business portfolio behind the number. We have written up a cause we found by living through it, which is a stale credit line shared into a portfolio by a provider the business no longer uses.

Sources