Skip to content

AI systems

WhatsApp chatbot or a person? What Meta allows, and how to hand a lead over

Some Indian businesses now answer WhatsApp enquiries with a bot, and some of their owners have heard that Meta has banned AI on WhatsApp. This guide sets out what WhatsApp’s own policy and terms say, where a person has to take over, and how to design that moment so the lead is not lost.

Where WhatsApp bots lose leads

Picture 9.40 on a Tuesday night. Someone in Baner has just got the keys to a flat and messages an interior design studio on WhatsApp. The bot replies in two seconds with a menu. She picks “2BHK”, gives her budget range and her area, then types: “Can someone call me tomorrow morning?” The bot sends the menu again. She asks once more and gets the menu a third time. By ten she has messaged two other studios.

The bot may even have been within WhatsApp’s written rules, if a phone number sat on the profile. It answered fast and collected useful details. It lost the lead at the one moment that mattered: when a person asked for a person. Most of what follows is about that moment.

Does WhatsApp allow chatbots?

Yes. The WhatsApp Business Messaging Policy (last updated 23 September 2026), in the section that applies to the WhatsApp Business Platform (the API), says: “You may use automation when responding during the 24-hour window, but must also have available prompt, clear, and direct escalation paths.” It then gives examples of paths:

  • an in-chat transfer to a human agent;
  • a phone number;
  • an email address;
  • web support on the business website;
  • an in-store visit, such as a retail store or bank branch;
  • a support form.

So the written rule does not demand a live person inside the chat. A phone number counts. The same policy also asks every business to “Maintain a WhatsApp Business profile with customer support contact information and one or more of the following: email address, website address, and/or telephone number.” A bot with no way out, on a profile with no contact details, fails both lines.

The 24-hour window is the other rule an API bot has to live with. The policy says you may reply freely while the window is open, which “opens and resets with each user message”, and that outside it “you may only send messages via approved Message Templates.” Meta’s developer documentation adds that a call from the customer also opens the window.

Has Meta banned AI chatbots on WhatsApp?

The fear comes from clause 4.7 of the Meta Terms for WhatsApp Business Platform (last modified 23 September 2026). It says providers of AI technologies, including “general-purpose artificial intelligence assistants”, are “strictly prohibited” from using the Platform to offer those technologies “when such technologies are the primary (rather than incidental or ancillary) functionality being made available for use, as determined by Meta in its sole discretion”. The same clause adds: “you may retain an AI Provider as your Solution Provider.”

Read plainly, the clause is aimed at companies whose product is the AI itself, such as a general assistant or an AI platform people use through WhatsApp. Meta’s page on AI providers (updated 15 October 2026) says such providers “are only permitted to offer general-purpose AI assistants on the WhatsApp Business Platform where Meta is legally required to permit this use case”. On that page the list is 32 markets: Brazil, Türkiye, Italy and 29 other European markets. India is not on it.

A bot that answers questions about your own furniture, courses or flats, and books a call, is a different thing. Meta’s pricing page treats it as ordinary business messaging: service messages “can be powered by a 3rd-party AI solution or people (like customer service representatives)”. That is our reading of the text, not a guarantee. Meta keeps the final word “in its sole discretion”, so ask any bot vendor how it fits the clause.

Two more lines in the same terms matter to anyone building a lead bot:

  • No training on your chats. Where you use an AI provider as your Solution Provider, WhatsApp Business Platform Data may not be used “to create, develop, train, or improve any machine learning or artificial intelligence systems”, except that you may use it to fine-tune “an AI Model that is for your exclusive use”. Put that in the vendor contract.
  • No profiling from WhatsApp data. Clause 4.1(a) bars using Platform Data “to track, build, or augment profiles on individual WhatsApp Users”, with an exception for “the content of your message threads”. Ask a lawyer how that applies to lead scoring in your CRM.

If an agency is reselling a WhatsApp AI service to you, our guide to white-label AI contracts covers the clauses to ask for.

The free app or the API: what each can automate

WhatsApp Business app

The free app’s built-in tools are not a custom bot, but three of them cover many small businesses:

  • A greeting message, sent on first contact and, per WhatsApp’s help page, “when customers message you after 14 days of inactivity”.
  • An away message, which can be set to “Outside of business hours” once your hours are on the business profile (help page).
  • Quick replies (“The maximum number of stored quick replies allowed is 50”, help page) and up to 20 labels, which WhatsApp is renaming lists (help page). A label such as “needs a call” is a perfectly good handoff queue.

Meta has also announced Business AI for small businesses in India in the app (May 2026). Meta says “a business owner can take over the conversation from the AI agent”, and that “Indian small and medium-sized businesses must meet certain criteria to be eligible”. Check in your own app whether it is offered to you.

WhatsApp Business Platform (the API)

A custom bot is built on the Platform, usually with a Business Solution Provider’s help. Meta now documents a handover protocol for it: one responder can pass a conversation to another, and “Only the account’s designated escalation partner can take control of a thread.” Three details from that page are worth giving to whoever builds your bot:

  • The handover event can carry a reason, such as “WhatsApp user requested human agent”, and in some cases an AI-generated summary of the conversation, so your staff do not start from hello.
  • “There is no endpoint that reports which responder currently owns a thread”, so the system has to track ownership itself.
  • If the wrong responder tries to reply, the message fails with an error saying “another app is controlling this thread now.”

It is not on by default. The page says standby visibility, the escalation partner and thread control “require the business to have accepted the Meta Business Agents and Platform Terms of Service”, and that “The account must be enrolled for thread control”. Ask your provider whether yours is.

Meta’s own Meta Business Agent hands over by itself “when it detects a signal such as low confidence, an integrity violation, or a customer asking for a human”, and you cannot change those triggers. Its eligibility page excludes some verticals, including finance, government and health; requires a business “based in a country authorized for Meta Business Agent”; and rules out a number that already has another AI agent connected. It does not list the countries, so confirm India with your provider.

Running a bot on the Platform has a cost. Since 1 October 2026 Meta charges for service messages beyond a free tier (and outside the free window that follows a click-to-WhatsApp ad), and an account with no payment method stops delivering them once that tier is used. Our WhatsApp Business API guide explains the pricing.

When should the bot hand over?

Nielsen Norman Group’s July 2026 article on AI chatbots says: “Never gatekeep when the user explicitly asks to be connected to a human. If a customer asks for a human agent, the bot should immediately honor that request.” It adds that a bot should also offer to escalate “when the user shows signs of frustration or extra effort (e.g., repeatedly rephrasing the same question)”.

TriggerWhat the bot should doWhy
The person asks for a human, a call or “someone”, in any languageHand over on the first ask. Confirm who will reply and by when.NN/g: honour the request immediately. The studio in Baner lost its lead here.
The same question twice, or the bot’s answer is not in its materialSay so and offer the person.NN/g lists repeated rephrasing as a sign to escalate. Guessing is worse.
Price negotiation, discounts, a complaint, a refundHand over with a short summary.These are decisions, not answers. Keep them with your team.
The bot would need an ID number, card or bank detailsStop and move to a phone call or a secure form.The policy says: “Don’t share or ask people to share” payment card, financial account or personal ID numbers.
The person says stopStop, and record the opt-out.The policy: “You must respect all requests ... to block, discontinue, or otherwise opt out”.

Test the first row yourself. Message your own number and type “human”, “call me”, “baat karni hai” and “kisi se baat karao”. If any of them gets a menu back, the bot is not finished.

What the person needs when the lead arrives

  1. The conversation, or a summary of it, attached to the alert. The lead has already told the bot the budget and the area; asking again tells her nobody read it.
  2. A clock. On the API, if the person replies after the 24-hour window has closed, they cannot write freely: outside the window the business can only send an approved template, and the window opens again when the customer next messages. Our guide to lead response time covers why the first hours matter anyway.
  3. A named owner. One person, not a shared inbox everyone assumes someone else is watching.
  4. A way back. Once the person is done, the bot can take routine questions again; on the Platform that means releasing the thread.

Tell people when they are talking to a bot. NN/g’s guidance is that AI chatbots should make clear “That they are an AI system”, and Meta does this with its own AI: its help page says such messages carry the marking “AI”, and that if you ask to stop AI-generated messages, it “will attempt to connect you with the business for additional support.” A third-party bot should do at least as much.

The Indian rules your bot’s script must follow

A lead bot collects personal data, so the Digital Personal Data Protection Act, 2023 applies once its main duties commence on 13 May 2027, the date the ministry gave in a May 2026 press release. A bot built now should already be ready for them:

  • Notice before the first data question. Section 5 says a consent request must be “accompanied or preceded by a notice” of the data and the purpose, how to withdraw and how to complain, and section 5(3) says the person must have the option to read the notice in English or any language of the Eighth Schedule. A bot that chats in Marathi should give its notice in Marathi too.
  • Ask only for what you need. Consent must be “limited to such personal data as is necessary for such specified purpose” (section 6(1)).
  • Easy withdrawal. Section 6(4) requires withdrawal to be as easy as giving consent: if a tap gave consent, a word such as STOP should withdraw it.
  • Keep proof. If consent is ever questioned in a proceeding, section 6(10) puts the burden on the business to prove that notice was given and consent obtained, so log what the bot showed and when.
  • Know your basis. Section 7(a) also lets a business use details a person volunteered for the purpose they gave them, such as answering the enquiry they sent, unless they tell you not to. Anything beyond that, such as marketing follow-ups, needs consent. Ask a lawyer where your line sits.
  • Your vendor is your responsibility. Section 8(1) keeps the business responsible for processing done on its behalf by a Data Processor, “irrespective of any agreement to the contrary”.

Our DPDP guide for small businesses covers the rest. The CCPA’s dark patterns guidelines also apply to “advertisers” and “sellers”, not only to platforms. Three of the listed patterns are easy for a bot script to fall into:

  • Nagging: “repeated and persistent interactions, in the form of requests” to push a transaction. A bot that asks for the phone number four times is close.
  • Forced action: making someone “share personal information in order to buy or subscribe to the product or service originally intended”. A bot that will not give a price range until it has a phone number is worth a second look.
  • False urgency: “falsely stating or implying the sense of urgency or scarcity”. “Only two site visits left this week” had better be true.

Whether a particular bot breaches these guidelines is untested. Write the script so the question never comes up. This is general information, not legal advice.

A checklist before your bot goes live

  1. The WhatsApp Business profile shows a phone number, email or website.
  2. The bot’s first message says it is automated and how to reach a person.
  3. Asking for a human, in English, Hindi, Marathi or Hinglish, gets a person on the first try, tested on your own phone.
  4. Repeated questions, complaints, negotiations and anything the bot cannot answer go to a named person with the conversation attached.
  5. On the API, someone watches the 24-hour window, and templates are approved for follow-ups after it.
  6. The bot never asks for ID, card or bank numbers in chat.
  7. A short notice comes before the first data question, in the customer’s language, with a way to withdraw.
  8. The bot asks for each detail once, and makes no scarcity claims you cannot back.
  9. The vendor contract bars training on your WhatsApp data and says how the vendor fits Meta’s AI-provider clause.
  10. On the API, a payment method is on the account and someone watches for policy warnings and user complaints, because users can block or report a business and Meta may limit the messages it can send.

QUESTIONS

Common questions

Are chatbots allowed on WhatsApp?

Yes. In its section for the WhatsApp Business Platform (the API), WhatsApp’s Business Messaging Policy says businesses may use automation to reply within the 24-hour window, as long as they have “prompt, clear, and direct escalation paths”, such as a transfer to a person, a phone number, an email or a support form.

Has Meta banned AI chatbots on WhatsApp in India?

Meta’s terms bar AI providers from offering AI as the “primary” functionality on the Platform; its AI-provider page lists 32 markets where general-purpose assistants are still allowed, and India is not one of them. Meta’s pricing page says ordinary service messages can be powered by a third-party AI. Our reading is that a business bot answering questions about its own products is ordinary service messaging, but Meta decides in its sole discretion, so ask your vendor how it fits the clause.

Do I need a live person inside the WhatsApp chat?

Not by the letter of the policy: a phone number, email, website support page, store visit or support form also counts as an escalation path. A person who can take over the chat is still the better design, and the bot should hand over the first time someone asks.

Can the free WhatsApp Business app run a chatbot?

The app offers greeting messages, away messages, up to 50 quick replies and up to 20 labels, not a custom bot. Meta has announced Business AI in the app for Indian small businesses that meet its criteria. Custom bots run on the WhatsApp Business Platform, usually through a provider.

Does a WhatsApp lead bot need consent under the DPDP Act?

When the Act’s main duties commence on 13 May 2027, a bot that relies on consent must give a notice of what it collects and why, ask only for what it needs, let people withdraw as easily as they agreed, and keep proof. Details a person volunteers to get an answer to their enquiry can generally be used for that purpose, unless they object; marketing beyond it needs consent. The business stays responsible for its bot vendor.

REFERENCES

Sources and further reading

Each link was opened and checked on .

← All field notes