Meta's WhatsApp AI rule: is your business chatbot allowed?

If you run a clinic, hotel or agency in India and an AI answers your WhatsApp enquiries, you may have seen headlines about Meta "banning AI chatbots" from WhatsApp. The plain answer: Meta banned general-purpose AI assistants offered by AI companies through the WhatsApp Business Platform, not the use of AI by an ordinary business to serve its own customers. But the wording leaves Meta a lot of discretion, so it pays to know exactly where the line sits.

What exactly did Meta ban?

Section 4.7 of Meta's terms for the WhatsApp Business Platform is titled "AI Providers". It says providers and developers of artificial intelligence or machine learning technologies are "strictly prohibited from accessing or using" the platform, directly or indirectly, when such technologies are the primary functionality being made available. It adds that Meta decides what counts as primary in its sole discretion.¹

TechCrunch reported the change on 18 October 2025: the ban applies to chatbots that are the main product, not to ancillary uses.³ It took effect on 15 January 2026, and OpenAI, Perplexity and Microsoft announced their WhatsApp chatbots would stop working.³

A Meta spokesperson told TechCrunch: "The purpose of the WhatsApp Business API is to help businesses provide customer support and send relevant updates."³ The article's own example of what stays allowed is a travel company running a customer-service bot.³

What stays allowed for a normal business?

Meta's terms say a business may keep an AI provider as its Solution Provider, meaning the company that connects the business to WhatsApp.¹ Meta's developer page also says its AI-provider charging policy "does not change charges for other businesses on the platform".²

Two data limits in the same terms are worth knowing:¹

• You may not let WhatsApp Business Platform data, even anonymous or aggregated, be used to create, develop, train or improve AI or machine-learning models.
• You may use that data to fine-tune an AI model for your own exclusive use, as long as it is not used to improve any other model.

How the rule evolved: a timeline

PPC Land says the Commission's decision applies across the European Economic Area, and that Italy's competition authority and Brazil's regulator took separate actions.⁴ Slaughter and May, the law firm advising Meta on the appeal, notes the June order was only the second time the Commission used this power under current EU competition law.⁵ The firm's page gives no outcome.

Does my clinic's or hotel's bot count as "primary AI"?

Meta has not published a test beyond "primary rather than incidental or ancillary", and says it judges this itself.¹ Reading the terms and the TechCrunch example together, the safest design is an assistant whose job is your business: your treatments, your rooms, your prices, your bookings.¹³

This table is our reading, not a Meta ruling. Meta decides.¹

A worked example: a dental clinic's WhatsApp assistant

A dental clinic in Pune lets an assistant answer enquiries on its number. A patient writes: "What does a root canal cost, and can you help with my history essay?"

• The cost question is the clinic's own business. The bot answers from the clinic's price list.
• The essay is off topic. A well-scoped bot says it can only help with the clinic's services and offers to book a visit.

That second behaviour is the point. A bot that happily writes essays is behaving like a general-purpose assistant, which is the thing Meta's rule is about.¹

Questions to ask your WhatsApp software vendor

Copy these into an email this week:

1. Is the AI on my number limited to my business's own services, prices and bookings?
2. What does it do when asked something off topic or something not in my knowledge?
3. Do you or your AI model provider use my customers' WhatsApp messages to train any model?
4. Are you, or is your AI model company, classed as an "AI Provider" under Meta's terms?
5. What is your plan if Meta changes the rule again?

Question 3 matters because the terms restrict using platform data to improve AI models.¹

What bananaleaf.ai does here

bananaleaf.ai is built as a sales assistant for your own WhatsApp number, not a general chatbot. Its replies are grounded in the business's own documents, approved facts and offerings sheet. When a question is not covered, it sends a holding line and flags your team instead of improvising. An off-topic message is acknowledged without being answered. Reply guards also cut invented prices, phone numbers and capacity claims. We have not been told by Meta how it classifies any product, so we cannot promise how Meta would judge a given setup; the design simply keeps the assistant on your business.

What to do this week

1. Write one sentence that says what your bot is for ("answers questions about our clinic and books visits").
2. Test off-topic questions from a spare phone and check the bot declines politely.
3. Send the five vendor questions above and keep the replies.
4. Check who your Solution Provider is. Meta's terms allow you to keep an AI provider in that role.¹
5. Do not rely on EU news for rights in India. Re-read Meta's AI-provider page for any India change.²
6. Diary a re-check for Meta's terms, last modified 23 September 2026 when we looked.¹