Updated on October 29, 2026
Since January 15, 2026, Meta’s terms prohibit AI Providers from using the WhatsApp Business API to offer general-purpose assistants, large language models or generative AI when that technology is the primary functionality of the service. Using AI on behalf of your own business is still allowed: answering customers, selling, scheduling and support. The rule lives in section 4.7 of the Meta Terms for WhatsApp Business Platform, and Brazil has a specific exception for AI Providers, with its own pricing.
This article summarizes what the official text says, what changed in practice, what falls outside the ban, and gives you a checklist to review your bot. It explains the rule but is not legal advice.
What the rule says (section 4.7)
The rule was introduced in the WhatsApp Business Solution Terms, the terms that govern the WhatsApp Business API. Today the old URL for those terms (whatsapp.com/legal/business-solution-terms) redirects to the Meta Terms for WhatsApp Business Platform, last modified on September 23, 2026. The topic is covered in section 4.7, titled “AI Providers”.
In short, the text:
- Defines who an AI Provider is: providers and developers of artificial intelligence or machine learning technologies, including large language models, generative AI platforms and general-purpose AI assistants, as Meta determines in its sole discretion.
- Bars them from the platform when those technologies are the “primary (rather than incidental or ancillary) functionality” of what is being offered. Under the terms, these providers are “strictly prohibited from accessing or using the WhatsApp Business Platform, whether directly or indirectly” for that purpose.
- Allows exceptions by country: the technology may be made available to businesses in certain countries, listed on a Meta developer page.
- Separates the vendor from the use: a business may keep an AI Provider as its Solution Provider. What it can’t do is let platform data train AI models (more on that below).
The key word is “primary”. The clause doesn’t ban AI on WhatsApp. It bans a conversation with a generic AI being the product delivered through the platform.
When it took effect
Meta’s official Pricing for AI Providers page states that the terms update took effect on January 15, 2026. From that date, AI Providers may only offer general-purpose AI assistants on the platform where Meta is legally required to permit it.
Pricing rules for those providers in the exception countries came next:
| Market | Start | Charges for the AI Provider’s service messages |
|---|---|---|
| Italy (+39) | Feb 16, 2026 | Charged Feb 16 to May 12, 2026; no charge since May 13, 2026 |
| Other EU countries, Iceland, Liechtenstein and Norway | Mar 11, 2026 | Charged Mar 11 to May 12, 2026; no charge since May 13, 2026 |
| Brazil (+55) | Mar 11, 2026 | Charged at a dedicated rate card, no published end date |
| Türkiye (+90) | Oct 15, 2026 | No charge |
Meta’s changelog entry of May 12, 2026 confirms charges ended in the European markets and that charges in Brazil remain. For the full timeline of the year, see every WhatsApp API change in 2026.
Want to try it while you read? Create a Vectalk account and send your first message through the official API. Create free account →
What was banned, in practice
The target is the “chat with an AI on WhatsApp” business model. Examples of services that fit the definition:
- an assistant that answers any question, on any topic, for the general public;
- a product whose value is access to an LLM through WhatsApp, unrelated to a specific company, store or service;
- a generative AI platform using a phone number as a distribution channel for its own model.
In these cases the rule applies both to whoever builds the model and to whoever packages it, because the text covers “direct or indirect” use. And Meta decides how a service is classified.
What’s still allowed
For most businesses, nothing changes. AI that works for the business is incidental or ancillary to the service the business provides. The WhatsApp Business Messaging Policy explicitly allows automation: you may use automated responses during the 24-hour window, as long as you have prompt, clear and direct escalation paths, such as an in-chat transfer to a human agent, a phone number, email or a support form.
| Scenario | How the rule reads |
|---|---|
| A store uses an LLM to answer questions about orders, returns and shipping | AI ancillary to the store’s support |
| A clinic uses an AI agent to book and confirm appointments | AI serving the scheduling process |
| An AI SDR qualifies leads and hands them to a sales rep | AI ancillary to selling the company’s products |
| A SaaS company uses AI to answer questions about its own software | AI ancillary to support |
| A service offers “ask an AI anything” on WhatsApp | AI Provider case; depends on the country exception |
If you’re building one of these flows, our guides on an AI chatbot for the WhatsApp API, connecting ChatGPT to the WhatsApp Business API and AI appointment scheduling on WhatsApp cover the technical side. For teams that serve customers after hours, see also 24/7 support with AI.
One caveat: scope. A store bot that accepts any question and answers about recipes, code or politics is, in practice, working as a generic assistant. Limit the topic to your business in the prompt and politely decline everything else.
Brazil’s exception for AI Providers
Brazil is on the list of exception countries. According to the AI Providers page, updated October 15, 2026:
- since March 11, 2026, AI Providers may send service messages to users with +55 numbers;
- those messages can only be sent in response to users, within the 24-hour customer service window;
- each delivered message is charged at the dedicated AI Provider rate card, which doesn’t apply to other businesses;
- in status webhooks, these messages carry the category
general_purpose_ai; in the Pricing Analytics API they show up asAI_BOT.
The relevant part of the pricing object in the webhook looks like this:
{
"pricing": {
"type": "regular",
"category": "general_purpose_ai"
}
}If your system calculates cost from webhooks, handle this category separately. For a regular business using AI in customer service, replies inside the window are still service messages: since October 1, 2026, each number gets 1,000 free service messages per month and pays BRL 0.0350 (Brazilian reais) from the 1,001st onward for recipients in Brazil.
Conversation data and model training
Section 4.7 also has a rule that applies to every business, not just AI Providers. A business may not, directly or indirectly, allow WhatsApp Business Platform data, including in anonymized, aggregated or derived form, to be used to build, develop, train or improve AI or machine learning systems or models.
There is one exception: a business may use that data to fine-tune an AI model for its own exclusive use, as long as the data isn’t used to train or improve any other model.
In practice, this affects your choice of LLM vendor. Before connecting a model’s API to WhatsApp, check that vendor’s terms on whether submitted data may be used for training, and opt out when that option exists. In Brazil, conversations are also personal data under the LGPD (Brazil’s data protection law, similar to the GDPR); our LGPD and WhatsApp article covers consent and opt-out.
Compliance checklist
Use this list to review an existing bot or a new project:
- Define the AI’s role. Write one sentence describing what the bot does for the business (support, sales, scheduling). If that sentence is “chat about anything”, you’re building an AI Provider product.
- Restrict scope in the prompt. The assistant answers about the company’s products, orders, services and calendar, and declines other topics.
- Offer a way out to a human. The policy requires clear escalation paths; the simplest one is a bot-to-human handoff.
- Respect the 24-hour window. Free-form AI replies only inside the window; outside it, approved templates only.
- Get opt-in and honor opt-outs. The policy requires permission to contact people and respecting requests to stop receiving messages.
- Review your LLM vendor’s contract. Make sure conversations aren’t used to train third-party models.
- Don’t reuse conversations for training. Only fine-tuning of a model exclusive to the business, and nothing that feeds other models.
- Monitor the pricing category. If
general_purpose_aishows up in your webhooks, your operation is being treated as an AI Provider. - Re-read section 4.7 whenever the terms change. The last modification was on September 23, 2026.
This checklist is for information only. If you’re unsure how your product is classified, talk to a lawyer and to Meta support.
How to do this with Vectalk
Vectalk is a proxy for Meta’s official WhatsApp Cloud API. It has no built-in AI in its dashboard: your AI agent runs in your own system or in tools connected to the API, receives messages through webhooks (in the same format as Meta’s documentation) and replies through the API. That keeps the bot’s scope, your LLM vendor and your data handling under your control.
In the dashboard you create the instance, set up the webhook, generate the API key and check the logs, including status events with their pricing category. Each delivered message is deducted from a prepaid balance, with no card registered with Meta. For background on the API, start with the official WhatsApp API guide; endpoints are at docs.vectalk.com.br and you can sign up at apps.vectalk.com.br/register.
Frequently asked questions
Can I use ChatGPT or another LLM for customer service on WhatsApp?
Yes, as long as the AI is ancillary to the business’s service, such as support, sales or scheduling. The ban in section 4.7 targets AI Providers offering general-purpose assistants as the primary functionality. Meta makes the final call on classification.
When did the rule take effect?
Meta’s official AI Providers page says the terms update took effect on January 15, 2026. The current version of the terms was modified on September 23, 2026 and keeps the rule in section 4.7.
Do general-purpose assistants still work in Brazil?
Brazil is one of the exception countries. Since March 11, 2026, AI Providers may send service messages to +55 numbers, only in response to users and within the 24-hour window, charged per message at a dedicated rate card.
Can I use WhatsApp conversations to train my model?
Not to build or train AI models in general. The terms only allow fine-tuning a model for the business’s exclusive use, without the data feeding any other model.
Do I have to tell customers they’re talking to an AI?
The WhatsApp Business Messaging Policy we reviewed has no specific requirement for that, but it does require a clear path to a human when you use automation. Saying it’s a virtual assistant is good transparency practice, and other laws may apply to your case.
Official sources consulted on October 29, 2026: Meta Terms for WhatsApp Business Platform, section 4.7 (last modified September 23, 2026); Pricing for AI Providers (updated October 15, 2026); WhatsApp Business Messaging Policy (updated September 23, 2026); WhatsApp Business Platform changelog.
Put your business on the official WhatsApp
Create your account, connect your number to the official API (you can keep the app with coexistence) and start serving customers and sending messages.