ChatGPT on WhatsApp — the ban, what Brazil and the EU decided, and what is still allowed

On January 15, 2026, ChatGPT stopped answering on WhatsApp in most of the world. The cause was a change to the WhatsApp Business Solution terms, the API businesses use to talk to customers: since that date, companies offering a general-purpose AI assistant can no longer use it to distribute that assistant. A business's own AI customer service bot is still allowed.

Brazil went a different way. Its antitrust authority, CADE, suspended the rule, Meta reopened access for +55 numbers while charging those providers US$0.0625 per message, and the fight over that fee still had no final ruling on September 24, 2026. Meanwhile, ChatGPT has not officially returned for Brazilian numbers: OpenAI announced its comeback only in the European Economic Area.

This piece is part of our WhatsApp news series: what Meta changes, with a date and a source, kept apart from rumor.

What Meta changed, and who left WhatsApp

The change was published on October 15, 2025 and reached a wide audience through a TechCrunch report three days later. The current WhatsApp Business Solution terms bar providers of AI technologies, such as large language models and general-purpose assistants, from using the platform to offer those technologies when they are the primary functionality, "rather than incidental or ancillary". Who decides what counts as primary? The text says Meta does, in its sole discretion.

According to TechCrunch, ChatGPT, Perplexity, Luzia, Poke and Microsoft's Copilot announced they would leave. ChatGPT alone had more than 50 million people using it on WhatsApp, according to OpenAI. Meta's argument was that the API was built for conversations between businesses and their customers, and that chatbots created a message volume the system was not designed to carry.

What Brazil's CADE decided

In November 2025, Luzia and Zapia took the case to CADE. On January 12, 2026, CADE's General Superintendence opened an inquiry into whether Meta uses WhatsApp's dominant position to favor Meta AI, and imposed an interim measure suspending the new terms. On the 15th, Meta told developers the ban no longer applied to numbers with the +55 country code.

From there, the case zigzagged. A federal court in Brasília suspended the measure on January 23. On March 4, CADE's Tribunal upheld it unanimously. Meta changed its terms again and, since March 11, has charged US$0.0625 for every message an AI provider sends to a Brazilian user, according to Meta's rate card for "AI Providers". In the Brazilian real version of the rate card, that is R$0.3217 per message.

For CADE, charging is just another way of keeping the door shut. On April 23, the Tribunal upheld a daily fine of R$250,000 for non-compliance: the order was to let chatbots resume their activities in the same way as before the change, with no extra charge. On April 30, a federal court in São Paulo suspended the fine and sent CADE and Meta into conciliation. As of September 24, 2026, we found no news of a settlement or a final decision, and Meta's rate card, updated on May 12, still lists Brazil with no end date for the fee.

What happened in Europe?

Italy came first. On December 24, 2025, the Italian competition authority ordered Meta to suspend the terms, and Meta began charging per message for +39 numbers on February 16. On June 8, the Italian authority closed its own case because the European Commission had taken over.

The Commission opened its investigation in December 2025. On May 13, 2026, Meta stopped charging AI providers in Europe. On June 9 came the tougher order: restore free access on the same terms as before October 15, 2025, within five working days, until the investigation ends. In its decision, the Commission wrote that the fee introduced in March was, at first sight, in practice equivalent to the ban. In July, ChatGPT came back to WhatsApp across the European Economic Area, and OpenAI's help page confirms the return in that region only.

Did anything change for WhatsApp users in Brazil?

For ordinary users, very little. Meta AI is still inside the app. ChatGPT, which left on January 15, has not announced a return for Brazilian numbers: OpenAI says availability depends on the country code and mentions only Europe. Whether any other assistant came back to serve +55 numbers by paying the fee, the sources we checked do not say.

On paper, the rule is clear: for Brazil, Meta's terms allow general-purpose AI providers. The open question is the price, and that is what CADE, Meta and the courts are still arguing about.

What is still allowed, and what is not

I run an AI customer service bot. Can I keep using it?

Yes. The ban targets companies that offer the assistant itself as the main product. A store whose bot answers questions about delivery times, or a clinic whose bot books appointments, is using AI to serve its own customers, which is exactly the use TechCrunch described as outside the rule.

What changes for that bot is the bill. Starting October 1, 2026, Meta will charge for service messages, the replies sent inside the 24-hour window, which have been free since November 2024. Businesses without a payment method on file by September 30 will stop having those messages delivered, according to Meta for Developers. Meta also launched its own AI agent for businesses on July 1, billed by usage since August 1.

How to tell whether your vendor is affected

Three questions settle most cases. Does the bot answer questions about your business, or about anything at all? Does the number belong to your company, or is it the vendor's number open to the general public? And does the contract, or the invoice, mention the fee for "AI Providers"? If the product is a generic assistant that serves anyone, it falls under the rule. If it is your own customer service with AI behind it, in principle it does not.

Since Meta has the final word on what counts as primary, get your vendor's answer in writing. And this is information, not legal advice.

What changes for people who serve customers on WhatsApp

A clinic front desk with a bot that confirms appointments carries on as before; it just needs to check October's new cost with whoever runs the system. The accountant who used ChatGPT on WhatsApp to draft replies to clients lost that shortcut in January and now has to open the ChatGPT app. The salesperson working from a phone feels no difference at all.

For a lawyer or a school, the point is different: any AI assistant that receives conversations from clients or families handles personal data, inside WhatsApp or out of it. And an agency that builds bots for clients needs to know which side of the line each project sits on: customer service for one brand is one thing, an assistant open to anyone is another.

If what your day is missing is a fast way to read incoming voice notes, not a robot talking to customers, there is a different category of tool. ZapVox, a WhatsApp Web extension that transcribes, summarizes and translates voice notes, works on the screen of your WhatsApp Web on a computer. It is not a number that talks to your customers, and it does not use the WhatsApp Business Solution: the text appears for you to read, and the reply is still yours.

Where this stops

• The case is not over. In Brazil, CADE's inquiry is still open and the fee depends on the court-ordered conciliation. In Europe, the June order stands until the Commission's final decision, which has no deadline.
• Meta defines "primary". The line between a customer service bot and a general-purpose assistant is written in the terms, but Meta is the one applying it.
• This is not legal advice. We describe public rules and decisions published up to September 24, 2026. If your business depends on this, talk to a lawyer.
• ZapVox does not reply for you. It works only on a computer, with WhatsApp Web open, has a daily quota on the Free plan and sends the audio to a cloud transcription service to turn it into text.