setscale

WhatsApp AI chatbot rules in 2026: what Meta allows

September 30, 2026·15 min read
Man around thirty with short curly black hair and a trimmed beard, wearing a navy knit polo, sitting at a round oak table in a small studio with plants and wooden shelves, phone in hand with the screen turned toward him, soft warm daylight from the side

Yes, a business can still run a WhatsApp AI chatbot in 2026. What Meta banned on 15 January 2026 is narrower than the headlines suggest: its terms stop "AI Providers" from offering a general-purpose assistant, the open, ask-me-anything kind, as the main thing they sell on the WhatsApp Business Platform, except in markets where Meta says it is legally required to allow it. An AI that answers your own leads, qualifies them and books a call sits on the other side of that line, and Meta told TechCrunch at the time that businesses using AI to serve their customers were not affected.

The line is drawn in documents that most "best chatbot" lists never quote. This update reads them directly, as they stand on 30 September 2026: the Meta Terms for WhatsApp Business Platform, section 4.7, last modified on 23 September 2026; the WhatsApp Business Messaging Policy, updated the same day; and Meta's developer page on pricing for AI Providers. Every rule below links to its source. None of it is legal advice.

TL;DR

  • Since 15 January 2026, Meta's terms bar AI Providers from the WhatsApp Business Platform when a general-purpose assistant is "the primary (rather than incidental or ancillary) functionality" on offer, as Meta determines. A bot that serves your own customers is not what the rule targets.
  • The per message charges Meta introduced for AI Providers in February 2026 do not apply to other businesses. For them, replies inside an open customer service window are still not charged.
  • Those AI Provider charges started in Italy on 16 February and in Brazil and a list of European countries, France included, on 11 March. The Italian and European ones ended on 12 May 2026. Brazil has no end date on Meta's page.
  • The Business Messaging Policy never mentions AI. Its one rule on automation: you may automate replies in the 24 hour window only with a prompt, clear and direct escalation path to a human.
  • Hiring a vendor's AI is explicitly allowed, but your WhatsApp data cannot improve a model beyond one that is for your exclusive use. For an agency, that is a per client rule.

Table of contents

What changed for WhatsApp AI chatbots, date by date

Most of the confusion comes from reading one headline from October 2025 and stopping there. The rule has moved several times since, and every step is dated in a primary document, so here is the sequence.

Date What changed Source
18 October 2025 TechCrunch reports a new "AI providers" section in the WhatsApp business terms, to take effect on 15 January 2026 TechCrunch
15 January 2026 The rule takes effect: AI Providers may offer general-purpose AI assistants only where Meta is legally required to permit it Meta developer page
16 February 2026 Meta starts charging AI Providers for each non-template message delivered to users in Italy Meta developer page
11 March 2026 The same charge extends to Brazil and to a list of European countries that includes France, Germany and Spain Meta developer page
12 May 2026 Last day of the charge in Italy and in the European list Meta developer page
23 September 2026 Latest revision of both the platform terms and the Business Messaging Policy Meta terms, WhatsApp policy

The update TechCrunch added on 15 January 2026 gives the context. OpenAI, Perplexity and Microsoft had announced that their WhatsApp chatbots would stop working, and regulators from the EU, Italy and Brazil had opened antitrust probes into the rules. Brazil, Italy and a list of European countries are also the markets Meta's page names as those where it is legally required to support AI Providers. Meta's page does not say why the Italian and European charges stopped on 12 May, and we will not guess.

Who Meta calls an AI Provider

The definition sits in section 4.7 of the Meta Terms for WhatsApp Business Platform, under "Platform Use". It covers "Providers and developers of artificial intelligence or machine learning technologies", followed by an open list that includes large language models and generative AI platforms and ends with "general-purpose artificial intelligence assistants, or similar technologies as determined by Meta in its sole discretion".

What the clause forbids is precise. AI Providers "are strictly prohibited from accessing or using the WhatsApp Business Platform, whether directly or indirectly" to provide, offer or sell those technologies "when such technologies are the primary (rather than incidental or ancillary) functionality being made available for use". The exception follows in the same clause: those technologies "may be made available to businesses in certain countries", with a link to the developer page on AI Provider pricing.

Read it as a buyer and three words do the work:

  • Primary: the AI is the product the user came for.
  • Incidental or ancillary: the AI serves something else the business does, such as selling a programme or a service.
  • Sole discretion: Meta decides which side you are on. Not your vendor, and not a blog post, this one included.

The WhatsApp AI chatbot a business can still run

Meta was explicit when the rule was announced. TechCrunch reported that Meta confirmed the move does not affect businesses using AI to serve customers on WhatsApp, and gave the example of a travel company running a bot for customer service. The company's spokesperson put the intent in one sentence:

"The purpose of the WhatsApp Business API is to help businesses provide customer support and send relevant updates. Our focus is on supporting the tens of thousands of businesses who are building these experiences on WhatsApp," a Meta spokesperson told TechCrunch.

Now apply that to someone who sells a service. A coach whose reel brings in a wave of WhatsApp messages wants an AI that answers each one, asks the few questions that tell a buyer from a browser, and offers a slot in the calendar. That AI serves the coach's own leads, for the coach's own offer. The same holds for a consultant, a training company, or an agency running that flow for each of its clients.

What moves you toward the other side is scope. A setter that answers questions about your offer, qualifies and books stays a business tool. The same bot, opened up to help whoever writes in with whatever they want, starts to look like a general assistant with your logo on it. Keep the job narrow in the prompt, not only in the pitch.

What the AI Provider charges mean for your bill

This is the second headline that worries buyers, and Meta's page answers it in bold. The pricing change for AI Providers "does NOT change how or what Meta charges all other businesses using the WhatsApp Business Platform". The next sentence matters even more for an AI chatbot that replies to inbound leads: "This includes not being charged for non-template messages sent in an open customer service window."

For a business that uses AI to answer its own leads, the price of a reply did not change. For AI Providers, Meta charges each non-template message delivered to a user in a covered market. Its example, written while Italy was covered, leaves no room for doubt: if a user in Italy sends a prompt and the provider answers with three messages within 5 minutes, that is three charges.

Market Charge to AI Providers per non-template message
Italy (+39) from 16 February 2026 until 12 May 2026
Brazil (+55) from 11 March 2026, no end date on the page
European list, including France (+33), Germany (+49), Spain (+34) from 11 March 2026 until 12 May 2026

The market list on the page was last updated as of 12 May 2026, and it states that from 13 May 2026 Meta no longer charges AI Providers for messages to users in the markets where the charge ended. The rates themselves are published as rate cards on the same page; we have not reproduced them here. For how the platform prices everything else, templates and windows included, see our WhatsApp Business API rules for teams.

Want your WhatsApp leads qualified and booked without hiring a setter? Start your 7-day free trial.

How to check which side of the line you are on

You do not have to infer your status from a sales page. Meta labels AI Provider traffic in two places that a developer on your side can read:

  • in status webhooks, the pricing object of a billable AI Provider message carries the category general_purpose_ai;
  • in the Pricing Analytics API, the pricing category AI_BOT reflects AI Provider traffic.

One caveat keeps this honest. Those labels sit on charged traffic, and today the only market on the page without an end date is Brazil. If you do not message users there, their absence proves little. Their presence, on the other hand, is unambiguous.

The better check happens before you sign. Three questions for any vendor selling you a WhatsApp AI chatbot:

  1. Is the bot scoped to our business, or is it a general assistant we are reselling under our name?
  2. Is our WhatsApp data used to train or improve any model other than one for our exclusive use?
  3. How does a person take over a conversation, and how fast?

The first maps to section 4.7, the second to the data rule further down in the same clause, and the third to the Business Messaging Policy.

The escalation path your AI chatbot must offer

Here is the rule most WhatsApp AI chatbot projects skip. The WhatsApp Business Messaging Policy, updated on 23 September 2026, does not contain the words AI or chatbot. Its one sentence about automation is this:

"You may use automation when responding during the 24-hour window, but must also have available prompt, clear, and direct escalation paths."

The policy then lists what counts as an escalation path: in-chat human agent transfer, a phone number, email, web support on the business website, in-store visits, or a support form.

For a business that sells calls, the in-chat transfer is the one that matters, and it is also the handoff you already need commercially. The moment a lead asks something the AI should not answer, about a refund, a payment plan or a custom deal, a person picks up the thread. Treat it as a written rule with an owner and a delay, the way you would write a lead response time SLA, and decide in advance who receives the conversation when the setter hands a qualified lead to a closer, as in any setter to closer handoff.

Your conversations cannot train a shared model

Section 4.7 does not stop at the AI Provider ban. It explicitly lets you use one: "Notwithstanding the foregoing, you may retain an AI Provider as your Solution Provider."

A condition follows. You may not let WhatsApp Business Platform data, including anonymous, aggregated or derived versions, be used to create, develop, train or improve AI models, with one exception: "you may use WhatsApp Business Platform Data to fine-tune an AI Model that is for your exclusive use", as long as that does not lead to the data being used for other models.

For a coach with one number, that is a vendor question: read your tool's data clause and keep a copy. For an agency running AI setting on several client numbers, it becomes an architecture question. Read literally, a model fine-tuned on one client's WhatsApp conversations has to stay for that client's exclusive use. Pooling several clients' conversations to improve one shared model is the pattern to check with counsel before you build it, in the same review where you settle who holds access to each client account.

Where every WhatsApp conversation has to start

An AI changes nothing about who you may message. The same policy says you may contact people only if "(a) they have given you their mobile phone number or username; and (b) you have received opt-in permission from the recipient confirming that they wish to receive subsequent messages or calls from you".

On the Business Platform, you "may only initiate conversations using an approved Message Template". Free-form replies, the kind an AI chatbot writes, are allowed "within the 24-hour customer service window, which opens and resets with each user message". Outside it, only approved templates go out.

In practice the AI does its work inside conversations the lead opened: a reply to a click to WhatsApp ad, a message from the link in your bio, an answer to a template the person agreed to receive. If you want to keep the WhatsApp Business app on the same number while the AI runs through the platform, the trade-offs are in our WhatsApp coexistence verdict.

Before you switch on an AI chatbot that books calls

One page, filled in once per number:

  1. Scope in one sentence. What the bot answers and what it hands over. If the sentence reads like "anything", narrow it.
  2. Escalation path. Pick one from the policy's list, name the person who owns it, and set the delay.
  3. Opt-in source for every entry point. Ad, link, form, template: each one has to produce the permission the policy describes.
  4. Window discipline. Free-form AI replies inside the 24 hour window; anything later goes through an approved template.
  5. Data clause in writing. No training beyond a model for your exclusive use, and for an agency, one per client.
  6. Brazil watch. If you message users in Brazil, have someone look for general_purpose_ai in your status webhooks.
  7. Revision check. Both the terms and the policy changed on 23 September 2026. The terms page links to its archived versions: compare them at each revision.

If you are wiring the same flow on Instagram, the equivalent limits and clocks are in our note on Instagram messaging API rate limits.

Where SetScale fits

SetScale is an AI setter in DMs, on Instagram, WhatsApp and Messenger, for people who sell a service. Every plan runs the full setter: it qualifies leads, books appointments and sends follow ups, from a single inbox. You start solo, on one account; when you grow, you add closers, accounts or client spaces without changing tools. The Scale and Business plans add the agency layer: separate client accounts, closers, reporting per client, white label and API.

See how it runs on the WhatsApp page, or read our guide to the AI appointment setter.

FAQ

Is it allowed to use an AI chatbot on WhatsApp Business in 2026? Meta's terms do not ban businesses from using AI to serve their own customers. Since 15 January 2026 they bar AI Providers from offering general-purpose assistants as the primary functionality on the WhatsApp Business Platform, except in markets where Meta is legally required to permit it. Meta decides which side a given use falls on, and local law still applies.

Can I put ChatGPT on my WhatsApp business number? Using a model from an AI company inside your own customer flow is allowed: the terms say you may retain an AI Provider as your Solution Provider. Offering an open assistant that answers anything on your number is the use the rule targets. The difference is the scope of the bot, not the brand of the model.

Does my WhatsApp chatbot have to say it is an AI? The Business Messaging Policy we read on 30 September 2026 does not contain the words AI or chatbot. It does forbid misleading customers about the nature of your business, and it requires an escalation path when you automate. The laws of the countries you sell in can add their own rules, so check them.

Can an agency run one AI setup for all its clients' WhatsApp numbers? Running the same tool is fine. Pooling data is the question. Section 4.7 allows fine-tuning on WhatsApp data only for a model that is for the business's exclusive use, so read literally, each client's conversations improve that client's model, not a shared one.

Conclusion

The ban that made headlines targets general-purpose assistants sold as the product. It does not target a business using AI to answer its own leads. What actually constrains a WhatsApp AI chatbot that books calls in September 2026 is more operational: a human escalation path, opt-in at every entry point, free-form replies inside the 24 hour window, and a data clause your vendor has to honour.

Put those four in writing before launch, and re-read section 4.7 each time Meta revises it. The next revision will not announce itself in your inbox.

Start your 7-day free trial