GUIDE  ·  10-Minute Read

WhatsApp AI Agent for Business: What It Can Do, Meta’s Rules, and How to Build One

By Zeeshan Nawaz

A WhatsApp AI agent answers your customers on WhatsApp, at any hour, in their language. It answers from your own business information, books appointments, qualifies leads and checks order status in your systems, and hands the chat to a person when it should. It is allowed on WhatsApp if the AI supports your business rather than being the product you sell. Meta’s terms now bar AI providers whose AI is the primary functionality. You also need the customer’s opt-in, must follow the 24-hour customer service window, and must offer a clear route to a human. This guide covers what such an agent can do, the rules, how it works, and how to roll one out.

Illustration of an agent hub calling out to seven connected tools
  • A WhatsApp AI agent connects a language model to the WhatsApp Business Platform, your business data and your systems, such as bookings, CRM and orders.
  • Meta’s terms bar AI providers from the platform when AI is the primary functionality being offered; a business using AI to serve its own customers is a different case.
  • You may only message people who opted in. Free-form replies are allowed for 24 hours after the customer’s last message; after that, only pre-approved templates.
  • Automation must come with prompt, clear and direct escalation paths, such as a transfer to a human agent.
  • The best first jobs are narrow and high-volume: order status, bookings, opening hours and FAQs, and lead questions before a salesperson calls.

What is a WhatsApp AI agent, and how is it different from a chatbot?

A WhatsApp AI agent understands free-text messages, answers from your business information and takes actions in your systems, such as booking a slot or checking an order. A classic chatbot only follows fixed menus.

A classic WhatsApp chatbot follows a menu: press 1 for opening hours, 2 for delivery areas. An AI agent understands free-text messages, including voice notes once they are transcribed, answers from your own business information, and takes actions through your systems. It can find a booking slot, check an order, or log a lead in your CRM, then hand over to a person when the conversation needs one. Sigi’s guide to AI agent use cases explains the difference between agents, chatbots and fixed workflows in more depth.

Risky

Menu chatbot

Fixed buttons and keywords. Breaks on anything unexpected, can’t look anything up, and sends frustrated customers back to waiting for staff.

Safer

Business AI agent

Understands normal messages, answers from your approved information, books or checks orders in your systems, and hands off to a person with the conversation summarised.

Are AI agents allowed on WhatsApp after Meta’s 2026 rules?

Yes, when AI supports your business rather than being the product. Meta’s terms prohibit AI providers whose AI is the primary functionality offered; Meta said businesses using AI for customer service are not affected.

Yes, for business use, with one important limit. Clause 4.7 of the Meta Terms for WhatsApp Business Platform strictly prohibits AI providers from accessing or using the platform when their AI technology is the primary functionality being offered, rather than incidental or ancillary. Meta decides that in its sole discretion. This is what removed general-purpose assistants from WhatsApp; TechCrunch reported in October 2025 that the change would take effect on 15 January 2026, and that Meta said it does not affect businesses using AI for customer service.

For a clinic, restaurant, retailer or logistics company, the AI supports a business that already exists, which is exactly the incidental use the clause describes. The same clause adds a data rule worth knowing. If your solution provider is an AI provider, WhatsApp Business Platform data cannot be used to train or improve AI models, except to fine-tune a model for your exclusive use. Ask any vendor how they handle that.

Which business problems can a WhatsApp AI agent solve?

Start where messages are frequent, repetitive and answerable from information you already have:

  • Order and delivery status: “Where is my order?” answered from your order system, for e-commerce, grocery, food delivery and courier businesses.
  • Appointments and reservations: booking, rescheduling and reminders for clinics, dental practices, salons and restaurants, written straight into your calendar or booking system.
  • Lead qualification: asking the questions your sales team asks first (location, timeline, requirements), common in real estate and B2B, so salespeople call only qualified leads, with the answers already in your CRM.
  • Quotes and catalogue questions: product availability and specifications for trade and wholesale businesses that already sell over the phone and WhatsApp.
  • Out-of-hours cover in two languages: answering in English and Arabic for Gulf customers, so evening and weekend messages are not left until the next working day. Test Arabic with real customer messages, because quality can vary between dialects.

What rules must a WhatsApp AI agent follow?

Get opt-in before you message people, reply freely only within 24 hours of the customer’s last message, give a clear route to a person, honour every opt-out, and follow personal data law.

  • Opt-in first. The WhatsApp Business Messaging Policy allows you to contact people only if you have their opt-in permission, and only if they gave you their number. You decide how to collect it, and you are responsible for it.
  • The 24-hour customer service window. Meta’s Cloud API guide explains that a customer’s message opens a 24-hour window, reset by each new message, during which you can send free-form service messages. When the window closes, you can only send pre-approved template messages. Your agent must know which mode it is in.
  • A route to a human. The same policy says automated experiences must have prompt, clear and direct escalation paths, such as a transfer to a human agent, a phone number, email or a support form.
  • Opt-out. You must respect every request to stop messages, whether it reaches you on WhatsApp or elsewhere.
  • Personal data law. Customer chats contain personal data. Check the law where your customers are, such as UK GDPR, Saudi Arabia’s Personal Data Protection Law and the UAE’s data protection laws.

How does a WhatsApp AI agent work behind the scenes?

Messages reach your server through the WhatsApp Cloud API, an AI model works out the request, answers from your data or acts through your systems, and replies through the same API, handing off to a person when needed. Step by step:

  1. The customer’s message arrives through the WhatsApp Business Platform (Cloud API) and is sent to your server by a webhook.
  2. Your server checks who the customer is, whether they opted in, and whether the 24-hour window is open.
  3. The language model reads the message together with relevant information retrieved from your FAQs, policies and catalogue. This retrieval approach is called RAG.
  4. If an action is needed, the agent calls your systems through tools you define: booking, order lookup, CRM. Each tool has limits on what it may change.
  5. Guardrails check the reply before it is sent: no invented information, no promises outside policy, and a handoff when confidence is low.
  6. The reply is sent, and the conversation is logged so the team can review quality and improve answers.

Retrieval keeps answers grounded in your own information; the details are in what is RAG.

Should you use a no-code WhatsApp AI builder or build a custom agent?

Use a no-code builder for FAQs and simple routing; build a custom agent when it must read and write your own systems, work in Arabic and English, or follow strict rules on what it may say.

  • A no-code builder or automation tool fits when: you need FAQs and simple routing, little connection to your own systems, and you accept the tool’s limits on data and behaviour.
  • A custom agent fits when: it must read and write your own systems (bookings, orders, CRM), work reliably in Arabic and English, follow strict rules on what it may say, keep data in a region you choose, or handle enough volume that quality monitoring matters.

Many businesses start with a narrow custom agent for one job, such as order status or bookings. They prove it with real customers, then add the next job. That is faster and safer than launching an agent that tries to answer everything.

How do you roll out a WhatsApp AI agent step by step?

Start with one or two frequent request types, test on real conversations, launch with people reviewing, then expand. Step by step:

  1. Export a month of real WhatsApp conversations and group them by type. Pick the one or two types that are frequent and answerable from your data.
  2. Write down the approved answers and policies the agent may use, and the topics it must always hand to a person.
  3. Set up the WhatsApp Business Platform with a verified business number, opt-in collection and template messages for anything sent outside the 24-hour window.
  4. Connect only the systems the first job needs, with read-only access wherever possible.
  5. Test with past conversations in English and, for Gulf customers, Arabic. Measure how many the agent answers correctly and how many it correctly hands off.
  6. Launch to a small share of customers with staff watching the handoff inbox, then widen as quality holds.
  7. Review conversations every week. Fix wrong answers at the source, in the information or the rules, not with one-off patches.

For which AI agent projects to fund first, read AI agent use cases and where they fail. For grounding answers in your own information, see what is RAG, and for a website chat assistant, how to build an AI chatbot. For phone calls, read AI receptionist for clinics and small businesses. For property enquiries, see AI agents for real estate lead qualification. The WhatsApp enquiry flows mentioned above are in the Madina Steel and UK dry cleaner case studies. Sigi’s AI agent development, RAG assistant and AI integration services cover this work.

Questions this guide answers

For businesses serving their own customers, yes. Meta’s terms bar AI providers whose AI is the primary functionality being offered, which removed general-purpose assistants from the platform from 15 January 2026. A business using AI to answer its own customers is using AI incidentally, the case the clause leaves out. Meta decides in its sole discretion.

Only customers who have opted in, and outside the 24-hour customer service window only with pre-approved template messages. Within 24 hours of the customer’s last message, the agent can reply with free-form messages.

Yes. The WhatsApp Business Messaging Policy requires prompt, clear and direct escalation paths for automated experiences, such as a transfer to a human agent, a phone number or email. Good design passes the conversation and a summary so the customer doesn’t repeat themselves.

Yes. Current language models can read and write Arabic and English, so one agent can serve both. Quality can vary between dialects and mixed Arabic-English messages, so test with real customer conversations before launch.

Yes, if it is connected to your booking or order system through tools with clear limits. Start with read-only actions such as order status, then add bookings once answers are reliable.

Yes. Sigi builds AI agents and RAG assistants that connect to your own systems, with a human handoff and conversation monitoring. We recommend starting with one high-volume job, such as order status or bookings, and expanding once it performs well with real customers.