WhatsApp AI Chatbots for Business: What Meta Allows

Full disclosure: we build WhatsApp CRMs, and the team behind Marketlube built Xale (opens in a new tab), whose WhatsApp inbox this guide uses as an example, so we sell part of what this guide discusses. We also tell small businesses when Meta's own tools in the WhatsApp Business app will do the job without us.

This guide covers what Meta's 2026 rules say, the three ways to put AI on a WhatsApp number, which enquiries belong with a bot or a person, the cost since 1 October 2026, and the data rules in India and the Gulf. It is our reading of public documents, checked on 4 October 2026, not legal advice.

What changed for AI chatbots on WhatsApp in 2026?#

On 15 October 2025 Meta announced new terms for businesses on the WhatsApp Business Platform, and they took full effect on 15 January 2026. The key clause is now section 4.7, "AI Providers", of the Meta Terms for WhatsApp Business Platform (opens in a new tab) (last modified 23 September 2026). It says providers of large language models and general-purpose AI 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", as Meta decides.

Read the clause closely and three things stay allowed:

Your own assistant for your own customers. A builder's bot that answers questions about its projects, or a college's bot that books counselling calls, uses AI as an incidental part of the business's service. The product is the flat or the course, not the AI.
An AI company as your provider. The same clause says "you may retain an AI Provider as your Solution Provider".
Fine-tuning for yourself only. Chat data from the platform may not be used to train or improve AI models, except to "fine-tune an AI Model that is for your exclusive use".

Meta's own pricing documentation backs this reading. Of service messages, the free-form replies a business sends inside a chat, it says: "These can be powered by a person, like a customer service representative, or by a 3rd-party AI solution" (Meta's pricing notes on service messages (opens in a new tab)).

Regulators pushed back on the ban itself. Italy's competition authority ordered Meta to suspend the terms (opens in a new tab) for Italy on 24 December 2025, and the European Commission sent Meta a statement of objections (opens in a new tab) on 9 February 2026, noting that since 15 January the only AI assistant available on WhatsApp was Meta's own. Meta's AI Providers page (opens in a new tab) charges AI providers per message in markets where Meta is legally required to let them in: Brazil since 11 March 2026 and, until 12 May 2026, the EEA and Italy. India is not on that list, so the ban applies here in full. It just doesn't touch a business's own assistant.

What can a business's own AI assistant on WhatsApp do?#

Inside the rules, an assistant can do most of the first-line work a sales or front-desk team does in chat:

Answer routine questions from content you approve: timings, location, fee structure, price list, documents needed, delivery areas.
Qualify the lead with the three or four questions your team always asks: budget band, location, timeline, which project or course.
Book the next step, such as a site visit, a counselling session, a test drive or an appointment, then confirm it with an approved template.
Report status from your CRM: where an application, order or booking stands.
Hand over to a person with the whole conversation attached, when the customer asks or the question needs judgement.

Two limits come straight from the WhatsApp Business Messaging Policy (opens in a new tab) (updated 23 September 2026). A bot must not ask people to share "full length individual payment card numbers, financial account numbers, personal ID card numbers, or other sensitive identifiers", so it never collects Aadhaar, passport or card numbers in chat. And health information is restricted where local rules require heightened handling, so a clinic's assistant should book appointments, not discuss symptoms.

Context helps before any AI does. Since 4 October 2026, every WhatsApp link on our own site opens a chat with a first message saying it came from marketlube.in and, where the page knows it, naming the service, article or page the visitor was reading. Whoever answers knows why the person is writing before typing a word. A pre-filled first message on your ads and pages does the same for a bot, and costs nothing.

Meta's own business agent, a platform bot or a custom assistant: which fits?#

There are three practical routes to AI on a WhatsApp number.

1. Meta's own agent in the WhatsApp Business app. Meta launched Business AI for small businesses in India (opens in a new tab) on 14 May 2026, "available in all native languages in India", trained on the business's own catalogue and documents, with the owner able to take over any chat. The sign-up page that announcement links to now presents it as Meta Business Agent (opens in a new tab), which learns from your past chats, website and price lists and lets you set when it hands off to you. Meta sells extra agent capacity through its Meta One subscriptions (opens in a new tab): business bundles start at $14.99 a month and include "more access to Meta Business Agent on WhatsApp" (15 September 2026). Plans and availability vary by region and account.

2. A bot from your WhatsApp provider or Meta's agent platform. On the API, many WhatsApp providers sell chatbot builders, and Meta's Business Agent Platform (opens in a new tab) "works alongside" the Business Platform for larger businesses (3 June 2026). Quick to set up, but the bot knows only what you load into it.

3. A custom assistant connected to your CRM. It answers from your live data (unit availability, intake dates, application status), follows your qualification rules and hands off into your own pipeline. It takes a build, which our guide to adding LLMs to business software explains, and it is the kind of build our AI automation service covers, with a person taking over whenever a chat needs judgement. On our builds the CRM underneath is Xale (reported by our team). Its own assistant, Xale AI (opens in a new tab), works on the team's side of the chat: staff ask it about their leads by chat or a live voice call in Malayalam or English, as xale.in describes it.

	Meta's agent in the app	Provider or platform bot	Custom assistant on your CRM
Set-up	Self-serve, inside the app	Configured in your provider's tools	Built as a software project
What it knows	Your catalogue, documents, chats	What you upload	Your live CRM data and rules
Hand-off	To the owner's phone	To agents in the provider's inbox	To the assigned person, inside your pipeline
Fits	One number, owner-run	Teams already on the API	Multi-branch teams with their own CRM

Once two or more people answer your enquiries, an agent inside the app runs into the app's own limits: one number on a few linked devices, and no assignment or response tracking. Our WhatsApp CRM guide covers the move to the platform.

Not every chat needs AI. A menu bot (buttons and fixed answers) is cheaper and never improvises; a person is slower but can judge. Match each kind of enquiry to the cheapest thing that answers it well:

Enquiry	Best first responder	Why
Timings, location, "send the brochure"	Menu bot	Fixed answers; buttons are faster than typing
"Is a higher-floor 3BHK still available?"	AI assistant reading live inventory	Needs current data, phrased naturally
"Which course suits my marks?"	Person, after the bot collects details	Advice with consequences for the student
"Book a site visit on Saturday"	AI assistant with the calendar	Routine, and confirmed by template
"Where is my application or order?"	AI assistant reading the CRM	Repetitive, answerable from records
Complaint, refund, failed payment	Person, immediately	Money and trust are at stake
Discount or negotiation	Person	Only a person can commit the business
Symptoms or medical questions	Person, by call	Rules on health data; risk of harm
An NRI writing at 1 am Gulf time	AI assistant, then a call at their hour	Captures the lead; the person follows up

The pattern: bots for facts, AI for routine questions that need your data, people for judgement, money and anything sensitive. For real estate sales teams and admissions offices, the most valuable bot job is usually the boring one: answer fast at night and book the call. Short of a bot, an after-hours template does part of that job: Xale's automations (opens in a new tab) can run a rule only outside business hours and once per lead, for example an after-hours WhatsApp reply to a new enquiry, as xale.in describes it.

What does a WhatsApp AI chatbot cost?#

Three costs stack up: Meta's message charges, the AI model, and the software around it.

Meta's charges changed on 1 October 2026. Service messages, the free-form replies a bot or a person sends, are now charged per message after a free tier of 1,000 a month for each business phone number, according to Meta's WhatsApp Business FAQ (opens in a new tab). Meta's notes on service message pricing (opens in a new tab) set each market's service rate at its utility and authentication rate. Meta's INR rate card effective 1 October 2026 charges ₹0.115 per service message to an Indian number, ₹1.1509 to a UAE number and ₹0.7834 to a Saudi number, plus GST.
Ads that click to WhatsApp open a free entry point window when you reply within 24 hours, and inside it marketing, utility, authentication and service messages are all free. The click-to-WhatsApp ads guide explains how to use it.
Meta Business Agent is priced by tokens on the API: $2 per million tokens, which Meta says typically works out at 20,000 to 25,000 tokens and about 4 to 5 US cents a message, about ₹3.84 to ₹4.80 at ₹95.99 to the dollar (the FBIL rate for 1 October 2026 on the Reserve Bank of India's website (opens in a new tab)). These messages are charged even inside the free entry point window.
A custom assistant pays its model provider by usage, plus your provider's or CRM's software fee and the build. Our AI automation cost guide lists what sellers in India charge for that build and its upkeep.

For scale, 5,000 bot replies a month to Indian numbers cost Meta's delivery charge of about ₹460 plus GST (4,000 messages after the free tier, at ₹0.115). The same 5,000 replies from Meta Business Agent would be roughly ₹19,000 to ₹24,000 at Meta's own per-message estimate. That gap is the budget a custom assistant's model costs have to fit inside. Our WhatsApp CRM guide holds the full pricing explainer, and the LLM integration guide works through model costs.

How should the hand-off to a person work?#

The policy sets the floor: you "may use automation when responding during the 24-hour window, but must also have available prompt, clear, and direct escalation paths", such as a transfer to a human agent, a phone number, email, web support, a store visit or a support form. A good hand-off goes further:

Escalate on clear triggers. The customer asks for a person, repeats a question, mentions a complaint or payment, or matches a high-value profile you define.
Pass the whole conversation. The person should see the transcript, the answers to the qualifying questions and the lead's source, so the customer never repeats themselves.
Assign to a named owner in the CRM, with a follow-up time, not to "whoever sees it". Xale, the CRM built by the team behind Marketlube, runs WhatsApp through the official Cloud API in a shared inbox scoped by role (opens in a new tab), so a chat sits with the people allowed to work that lead rather than on one shared phone.
Beat the 24-hour clock. Free-form replies are allowed only within 24 hours of the customer's last message; after that, only approved templates. Escalations that land at 7 pm need someone, or a template, ready before the window closes.
Say it's an assistant. Telling people they are chatting with an assistant, and how to reach a person, sets fair expectations from the first message.

What rules apply in India and the Gulf?#

This section summarises public rules; ask a lawyer before you process health, financial or children's data.

Opt-in, on any market. The policy allows you to contact people only if they gave you their number and "opt-in permission ... to receive subsequent messages or calls from you", and you must honour opt-out requests made on or off WhatsApp.
India. The Digital Personal Data Protection Rules were published in the Gazette on 13 November 2025, and their duties on consent notices, security and breach reporting apply eighteen months later (opens in a new tab), in May 2027. Rule 3 (opens in a new tab) asks for an itemised description of the data collected and its purpose, and a way to withdraw consent as easily as it was given. A chatbot that asks for a name, budget and city is collecting personal data, so the notice applies to it. Our DPDP checklist for leads and CRM data covers each touchpoint.
Your AI vendor's terms. Meta's clause keeps platform chat data out of other companies' models. Check that your model provider and chatbot vendor don't train on your customers' chats either.
UAE. Federal Decree-Law No. 45 of 2021 is the national data protection law, with separate regimes in the DIFC and ADGM free zones. Its implementing regulations "have yet to be issued", according to Chambers' 2026 UAE guide (opens in a new tab) (updated 10 March 2026).
Saudi Arabia. The Personal Data Protection Law came into force on 14 September 2023, with a grace period to 14 September 2024, according to SDAIA's guide to the law (opens in a new tab). The guide also notes that sensitive data may not be processed for marketing, even with consent.

A launch checklist#

Before your assistant answers its first real customer:

The use case is your own service, not a general-purpose assistant
Every contact it messages first has opted in, and opt-outs are recorded
Answers come only from content you approved, with prices and dates kept current
It never asks for card, bank or ID numbers, and never gives medical advice
It tells customers they are talking to an assistant and how to reach a person
Escalation triggers are defined and tested, with a named owner for each
The transcript, answers and lead source land in your CRM
Templates are approved for follow-ups after the 24-hour window
Your model and chatbot vendors confirm in writing they don't train on your chats
Someone reviews twenty bot conversations a week for the first month

Frequently asked questions#

Is it allowed to use an AI chatbot on WhatsApp Business?#

Yes, for a business answering its own customers. Meta's 2026 terms prohibit AI providers from offering general-purpose AI assistants on the platform where the AI is the main product, except in markets where Meta is legally required to permit it. A business's own support, booking or sales assistant uses AI as an incidental part of its service, and Meta's pricing documentation explicitly says service messages can be powered by a third-party AI solution.

Can I connect ChatGPT or another AI model to my business WhatsApp?#

You can use a model from OpenAI, Anthropic, Google or another provider to power your own business's assistant, usually through your WhatsApp provider or a custom integration. What you cannot do is offer a general-purpose assistant to the public as the product. Chat data from the platform may only be used to fine-tune a model for your exclusive use, not to train anyone else's.

How much does a WhatsApp AI chatbot cost in India?#

Since 1 October 2026, Meta's FAQ says the first 1,000 service messages each month are free for every business phone number; after that, Meta's INR rate card charges ₹0.115 per service message to Indian numbers, plus GST. Meta's own Business Agent is priced at $2 per million tokens on the API, about 4 to 5 US cents a message by Meta's estimate. Add your provider's software fee, and model costs if you run your own assistant.

Does a WhatsApp AI chatbot work in Malayalam?#

Meta says its Business AI for Indian small businesses is available in all native languages in India. For a custom assistant it depends on the model you use, so test it with real customer messages, including Malayalam typed in English letters, before launch.

Do I need the WhatsApp Business API to use an AI chatbot?#

Not for Meta's own agent, which runs inside the WhatsApp Business app for eligible businesses. You need the Business Platform (the API) for a provider's bot or a custom assistant connected to your CRM, and for several people to answer from one number.

What to do next#

List your last hundred WhatsApp enquiries by type and mark which ones a bot could have answered from facts you already publish. That tells you whether a menu bot, Meta's agent or a custom assistant fits. If you want help deciding, send us a few sample chats and we will tell you honestly which route suits your volume. If AI in your wider sales process is the bigger question, start with our guide to AI in CRM.

Changelog#

2026-10-11: Linked our AI automation cost guide from "What does a WhatsApp AI chatbot cost?", for what a custom assistant's build and upkeep cost.
2026-10-12: Linked xale.in where this guide already uses Xale as its example: the platform page from the disclosure and the WhatsApp inbox page from the hand-off step. The custom assistant route gains two sentences: the CRM under our builds is Xale (reported by our team), and Xale AI works on the team's side of the chat, as xale.in describes it. The bot-or-person section gains one sentence on after-hours templates sent by Xale's automations, with a link to its automation page.