Under Meta's 2026 WhatsApp rules, a business can still run an AI chatbot that answers its own customers about its own products, orders, bookings and support. What Meta prohibits is using the WhatsApp Business Platform to distribute a general-purpose AI assistant, such as a ChatGPT-style bot, when the AI itself is the main product being offered. That ban took effect for all existing users on 15 January 2026, with exceptions forced by regulators in a few markets.
What changed in Meta's WhatsApp terms
In October 2025 Meta added an "AI Providers" clause to the terms that govern the WhatsApp Business Platform (formerly called the WhatsApp Business Solution Terms). TechCrunch reported that the new terms would take effect on 15 January 2026 and would likely affect WhatsApp-based assistants from OpenAI, Perplexity, Luzia and Poke. By the January deadline, TechCrunch's update notes that OpenAI, Perplexity and Microsoft had announced their WhatsApp chatbots would stop working.
The clause now sits in Section 4.7 of the Meta Terms for WhatsApp Business Platform (the version we reviewed was last modified 23 September 2026). It defines AI Providers as providers and developers of AI or machine learning technologies, "including but not limited to large language models, generative artificial intelligence platforms, general-purpose artificial intelligence assistants, or similar technologies as determined by Meta in its sole discretion." Those providers are prohibited from using the platform to make such technologies available "when such technologies are the primary (rather than incidental or ancillary) functionality being made available for use."
Meta's stated reason, given to TechCrunch, was that "the purpose of the WhatsApp Business API is to help businesses provide customer support and send relevant updates," and that general-purpose chatbots created message volumes and support needs the company was not ready for.
The key test: primary versus incidental
The whole policy turns on one distinction. Meta does not ban AI on WhatsApp; it bans AI as the primary product delivered through the Business Platform. TechCrunch's reporting gives the example that a travel company running a customer-support bot is not restricted. In practice, ask yourself: would customers still message this number if the AI disappeared and a human answered instead? If yes, because they want your hotel room, your parcel or your refund, the AI is incidental. If the only reason the number exists is to give people access to a chatbot, you are likely in AI Provider territory.
Note that Meta reserves the final call "in its sole discretion" twice in the same sentence. There is no published checklist of borderline cases, so treat grey areas conservatively.
Allowed vs not allowed: a practical table
| Use case | Status under the 2026 terms | Why |
|---|---|---|
| AI agent answering customer support questions and FAQs for your own business | Allowed | AI is incidental to the service you provide |
| AI handling order tracking, returns and delivery questions for your store | Allowed | Supports your commerce, not a standalone AI product |
| AI booking appointments, tables or rooms for your clinic, restaurant or hotel | Allowed | Booking is the service; AI is the interface |
| AI qualifying sales leads and answering product or pricing questions | Allowed | Sales of your own offering is the primary purpose |
| Hiring an AI company as your Solution Provider (BSP) to build your bot | Allowed, with data limits | Section 4.7 says you "may retain an AI Provider as your Solution Provider," but restricts use of your WhatsApp data for training other AI models |
| A ChatGPT-style assistant on a WhatsApp number that answers anything for anyone | Not allowed (outside regulator-exempt markets) | General-purpose AI is the primary functionality |
| An AI start-up distributing its LLM product to consumers through the Business API | Not allowed (outside regulator-exempt markets) | Matches the "AI Providers" definition directly |
| Feeding WhatsApp chat data into a model that a vendor reuses for other customers | Not allowed | Data may only fine-tune an AI model "for your exclusive use" |
The data rule most businesses miss
Section 4.7 also contains a data restriction that is easy to overlook. If you retain an AI Provider as your Solution Provider, you may not allow WhatsApp Business Platform Data, including "anonymous, aggregate, or derived forms," to be used to "create, develop, train, or improve" any AI models. The one carve-out is that you may use that data "to fine-tune an AI Model that is for your exclusive use," as long as it does not end up improving any other model.
In plain terms: before you connect an AI vendor to your WhatsApp number, check its data-processing terms. If the vendor pools customer conversations to improve a shared model, that conflicts with the wording of Meta's terms. Ask for this in writing.
Where the ban does not fully apply: EU, Italy and Brazil
The policy has been challenged by competition regulators, and the result is a patchwork. Here is what we could confirm from the sources we read.
Get the DPDP WhatsApp checklist
A founder-led WhatsApp reply with the DPDP consent + audit-log checklist for WhatsApp Business messaging. India-hosted. No spam.
Brazil
On 13 January 2026, two days before the deadline, Brazil's competition authority CADE ordered Meta to suspend the policy and opened an investigation, according to TechCrunch. CADE said there was "possible anti-competitive conduct of an exclusive nature." Meta then changed its terms to charge for messages sent by AI chatbots in Brazil. CADE's own announcement says its tribunal unanimously upheld a violation notice over this, finding Meta had sought to classify these players "as marketing messages, subject to per-message charges," and kept a daily fine of BRL 250,000 in place until full compliance.
European Union and Italy
TechCrunch reported that the EU and Italy were also investigating, and that Meta allowed AI providers to keep operating in Italy after 15 January. The European Commission opened its investigation in December 2025. According to the Law Society Gazette and a Houthoff analysis, Meta readmitted rival AI assistants in March 2026 but with fees, and on 9 June 2026 the Commission imposed interim measures ordering Meta to restore free access for rival general-purpose AI assistants on pre-October 2025 terms while its investigation continues. Commissioner Teresa Ribera said: "In rapidly evolving markets, competition can be lost long before a final decision is adopted."
What Meta's own pages say now
Section 4.7 now states that AI Provider technologies "may be made available to businesses in certain countries" and links to Meta's AI Provider pricing page. That page lists Brazil, Italy and a set of EEA countries with specific effective dates during 2026, and says that from 13 May 2026 Meta "will no longer charge 'AI Providers' for non-template messages delivered to users in certain markets." We could not fully reconcile the exact current charging status for each market from the pages alone, and these cases are still live. If you are an AI company targeting the EEA or Brazil, read that page and take legal advice rather than relying on any summary, including this one.
The markets this article is written for, Singapore, the UAE, the UK and the US, did not appear on the country list we reviewed. On that reading, the default rule applies there: general-purpose AI assistants as a primary product are prohibited, and business AI that is incidental to your service is fine.
Rules that still apply to every WhatsApp AI chatbot
Passing the AI Provider test does not exempt you from the rest of Meta's rulebook. The WhatsApp Business Policy (last updated 23 September 2026) has three requirements that matter most for automated agents.
1. A route to a human
The policy says: "You may use automation when responding during the 24-hour window, but must also have available prompt, clear, and direct escalation paths." The options it lists include in-chat human agent transfer, phone, email, web support, in-store visits or a support form. An AI agent that loops forever with no way out is a policy problem, not just a customer-experience problem.
2. Opt-in before you message
You may only contact people who have given you their number and from whom you have "received opt-in permission" confirming they want messages from you. An AI agent that replies to inbound chats is fine; an AI agent that starts cold outreach to scraped numbers is not.
3. Honour every opt-out
Businesses must "respect all requests (either on or off WhatsApp) by a person to block, discontinue, or otherwise opt out of communications." For an AI agent this means detecting "stop" or "unsubscribe" in whatever language the customer writes, and not letting the model talk them out of it.
A compliance checklist for businesses using AI on WhatsApp
- Scope the bot to your business. Ground answers in your own catalogue, policies and order data. Politely decline unrelated requests ("write my essay", "who won the match") rather than acting as an open assistant.
- Build the human handoff first. Make transfer to a person available on request and whenever the bot is unsure.
- Check your vendor's data terms. Confirm your WhatsApp conversations are not used to train shared models.
- Keep opt-in records and process opt-outs in every language your customers use.
- Watch costs as well as rules. AI replies inside the customer service window and template messages are priced differently by country; see our WhatsApp API pricing by country guide or run the numbers in the WhatsApp pricing calculator by country.
- Re-check the terms periodically. Section 4.7 has changed more than once since October 2025 and regulators are still involved.
How RichAutomate fits the 2026 rules
RichAutomate is built on Meta's official Cloud API and is designed for the use case Meta explicitly allows: a business answering its own customers. The AI agent handles support, FAQs, orders and bookings for your business and replies in the customer's language. When a conversation needs a person, it hands off to your team inbox, which covers the escalation requirement. You can shape the conversation with the visual no-code flow builder, and opt-out keywords are handled in multiple languages so "stop" means stop.
If you are planning AI-led campaigns as well as support, our 2026 WhatsApp campaign read-rate study shows what to expect, and RichAutomate international pricing lists plans for businesses outside India.
This article summarises public sources as of October 2026 and is not legal advice. Meta's terms and regulators' orders change; check the official pages linked above before making decisions.