AI agent on WhatsApp Business: what it can do, what Meta allows and how to set it up

An AI agent on WhatsApp Business can answer frequent questions, qualify leads, book appointments and hand the conversation to a person when needed, around the clock. To do this automatically you need the WhatsApp Business Platform (Cloud API), not just the app, and the assistant must stay focused on your company's services: since 15 January 2026 Meta's terms restrict general-purpose AI assistants, but not customer-service assistants.

This guide explains what such an agent can do, what it should not do, what Meta's policy says today, what EU rules require and how to implement one step by step in an SME, a professional firm, an estate agency or a clinic operating in Spain.

What an AI agent on WhatsApp Business can do

A well-designed AI agent is not a menu bot with buttons. It understands messages written in natural language, looks up your company's information and acts within defined limits. Four tasks usually deliver the most value.

Answering frequent questions

Opening hours, address, documents needed for a procedure, services offered, usual timeframes. The agent answers from a knowledge base that you control and update, not from whatever the model "knows" in general.

Qualifying leads

Before a customer speaks to your team, the agent can collect the basics: what they need, by when, where, and whether they are an individual or a company. An estate agency, for example, can ask about approximate budget, property type and area before offering a viewing.

Booking and managing appointments

Connected to your calendar, the agent can offer available slots, confirm bookings, send reminders and handle changes. It is one of the easiest uses to measure in clinics and professional firms.

Handing over to a person

This matters as much as the rest. When a query is complex or sensitive, or the customer asks for it, the agent should pass the conversation to your team with a summary. WhatsApp's Business Messaging Policy requires automated experiences to offer prompt, clear and direct escalation paths, such as transfer to a human agent, a phone number or an email address.

What an AI agent should not do

Setting limits is part of the design, not an afterthought:

• Give definitive professional advice. A law or tax firm can use the agent to guide and collect information, but legal, tax or medical opinions must come from a professional.
• Make up answers. If the information is not in its knowledge base, the agent should say so and offer to put the customer in touch with a person.
• Process data it does not need. Asking only for what is strictly necessary reduces risk and simplifies GDPR compliance.
• Commit to prices, deadlines or terms you have not approved.
• Act as a general assistant. Besides being of little use to your business, an assistant open to any topic is exactly the use Meta has restricted.

WhatsApp Business app or WhatsApp Business Platform

The choice depends on volume and on what you want to automate.

Many companies do not know that Meta allows the same number to be used on the app and the Cloud API at the same time (what it calls "coexistence"), with message history synced between both. Your team can keep answering from the app while the agent works through the API. There are limitations: for instance, broadcast lists and group chats are no longer available on that number.

What Meta's policy on AI allows today

In October 2025 Meta changed the WhatsApp Business Solution terms to prohibit "AI providers" from using the platform when AI is the primary functionality they offer. The change applied to new accounts from 15 October 2025 and to all existing ones from 15 January 2026. Its target was general-purpose assistants of the "chat with an AI about anything" kind.

Meta said at the time that the measure does not affect businesses using AI to serve their own customers, for example an assistant that handles bookings or support requests. That is the case of an SME or a professional firm: AI is a tool for its customer service, not the product it sells.

In the European Economic Area the situation has moved on. The European Commission opened a competition investigation and, on 9 June 2026, imposed interim measures ordering Meta to restore free access for rival general-purpose AI assistants on the terms in place before October 2025, for as long as the investigation lasts. Meta announced it would challenge the decision.

In practice, the conclusion for your company does not change: an assistant focused on your services, customers and appointments is a permitted use of the platform. It is still worth checking the terms in force when you sign up with a provider, as the matter remains open.

Opt-in, the 24-hour window and templates

Three WhatsApp rules shape how the agent is designed:

1. Opt-in. You may only message people who have given you their number and agreed to receive your messages. The opt-in must make your business name and the purpose of your messages clear.
2. The 24-hour customer service window. Every time a customer messages or calls you, a 24-hour window opens, and it resets with each new message from them. Within it you can reply with free-form messages, including those drafted by the agent.
3. Templates. Outside that window you can only send templates pre-approved by Meta, classified as marketing, utility or authentication. An appointment reminder, for example, is sent as a utility template.

This means the AI agent mainly works in conversations the customer starts. Proactive messages, such as reminders or notices, are prepared as templates.

Keep the platform's cost model in mind too. Meta charges per message and, according to its documentation, from 1 October 2026 service messages sent within the 24-hour window, which were previously free, are also charged. Meta also offers its own agent, Meta Business Agent, billed on usage since August 2026. Check current rates before estimating running costs.

GDPR and the EU AI Act

A WhatsApp agent processes personal data: name, phone number and whatever the customer tells it. The GDPR therefore applies. The basics are:

• A clear legal basis for each processing activity (for example, handling the customer's request).
• Information for users about who processes their data and why, accessible from the conversation.
• Data processing agreements with the providers involved (messaging platform, AI model provider, hosting).
• Control over international data transfers.
• Defined retention periods and, if you handle special-category data such as health data in a clinic, a more rigorous risk assessment.

In addition, Article 50 of Regulation (EU) 2024/1689 on artificial intelligence, applicable since 2 August 2026, requires that people know they are interacting with an AI system unless it is obvious. A short notice at the start ("I am the virtual assistant of…, I can help you with… and put you through to a person whenever you need") covers this and builds trust.

How to implement an AI agent on WhatsApp: steps

1. Choose a single use case. For example, booking appointments or answering the twenty most frequent questions. Not everything at once.
2. Measure the current situation. Messages per month, time to first reply and hours your team spends on them.
3. Prepare the knowledge base. Approved answers, opening hours, services, documents, and which topics must always go to a person.
4. Set up the WhatsApp Business Platform. A Meta Business account in your company's name, a verified number and, if you wish, coexistence with the app.
5. Connect the systems you need. Calendar, CRM or contact form, with minimum permissions.
6. Define human handover. What triggers it, who receives it and how the customer is told.
7. Review GDPR and transparency. AI notice, privacy information and agreements with providers.
8. Create and get approval for templates for reminders and notices outside the 24-hour window.
9. Test with real conversations in a limited pilot before opening it to all customers.
10. Measure and adjust every week during the first month.

How to measure results

Use the same indicators before and after the pilot:

• Time to first reply, especially outside office hours.
• Resolution rate without human intervention: conversations the agent closes correctly.
• Handover rate: if it is very high, the agent lacks information; if very low, it may be answering things it should not.
• Business results: appointments booked, qualified leads, viewings scheduled.
• Quality: a weekly review of a sample of conversations and of any complaints received.

Without a baseline you cannot tell whether the agent improves anything. Record your starting figures before you begin.

How we work at BL Diseno

At BL Diseno, a technical studio in Barcelona, we build WhatsApp AI agents around one concrete, measurable use case, with handover to your team from day one. Accounts and configuration stay in your company's name, the price is agreed in writing before we start, and you deal with a single point of contact.