How does a rule-based WhatsApp chatbot work?
A rule-based chatbot, also called a menu bot, follows a decision tree someone drew in advance. The customer taps a button or types a keyword, and the bot sends the linked reply. It does not read the sentence. It checks it against its rules.
WhatsApp supports this directly, with reply buttons (up to three predefined replies, per Meta’s developer documentation) and list messages with a set of options.
A menu bot says only what you wrote, so it cannot invent a price. It suits short, stable answers such as opening hours. But when the customer writes something the tree did not expect, it repeats the menu, and every new question needs a new path drawn by hand.
What is an AI sales agent, and what does it do differently?
An AI sales agent reads the customer’s words and decides the next step: answer, ask, book or hand over. On your official WhatsApp Business API number, it:
- understands free text in Arabic (the dialect you ask for) and English, at any hour;
- answers from a list of your prices, timings, locations and policies that you signed off;
- asks the 2 or 3 questions your team always asks, and saves the answers in the CRM;
- books a slot, or sends your payment link in the chat;
- hands over to a person, with the whole chat, when the question needs one.
It can still send buttons. What changes is who decides the next step: a tree drawn in advance, or the customer’s words. It does not replace your team: negotiation, custom quotes and complaints stay with your people.
What does the same question look like in each?
Here is one question, sent at 21:40 to a menu chatbot and to an AI sales agent. Both chats are examples we wrote, not real customers. A better menu could add a “weekend classes” button, but no menu has a path for every next question.
A swimming academy in SharjahMenu chatbot · example
- Parent: Hi, do you have a weekend class for my 6-year-old? We live in Al Khan.
- Menu chatbot: Welcome! Reply with a number: 1 Classes, 2 Fees, 3 Locations, 4 Contact us
- Parent: I asked about a Saturday class for a 6-year-old
- Menu chatbot: Sorry, I didn’t understand. Reply with a number: 1 Classes, 2 Fees, 3 Locations, 4 Contact us
- The chat ends here: no booking, no answers saved, and nobody on the team has seen it.
A swimming academy in SharjahAI sales agent · example
- Parent: Hi, do you have a weekend class for my 6-year-old? We live in Al Khan.
- AI agent: Hi! I’m the academy’s AI assistant. Yes: Saturday classes for ages 5 to 7, at 09:00 and 11:00. Can he or she swim yet?
- Parent: No, a complete beginner. A boy.
- AI agent: The beginners’ group fits him. 09:00 or 11:00? I can book a trial class.
- Parent: 11:00. He has mild asthma, is the pool OK?
- AI agent: Booked: Saturday 11:00, trial class. The asthma question is for a person, so I’ve passed this chat to our head coach, who will call you tomorrow morning.
- CRM: trial class booked, child’s age and area saved, chat sent to the head coach, call set for 09:30.
The menu bot did what it was built to do. It was not built for this parent.
What can each one do, and what can it not?
| When… | Rule-based chatbot | AI sales agent |
|---|---|---|
| The customer writes | Matches a button or keyword | Reads free text, dialect and mixed Arabic and English |
| It needs an answer | Sends the fixed reply for that path | Answers from your approved list |
| The question was not planned | Repeats the menu | Answers if the list has it; if not, says a colleague will reply and passes the chat on |
| You need to qualify the buyer | A fixed form, one field per step | 2 or 3 questions inside the chat, saved to the CRM |
| The buyer is ready | Sends a link to a booking page | Books a slot or sends your payment link in the chat |
| Setup | You draw every path; each new question is a new path | Answer list signed off by you; live in 2 weeks |
| Main risk | Dead ends: a customer who leaves the menu gets nowhere | A wrong answer if the list is thin or old, which the controls below limit |
| It fits when | Customers ask a handful of stable things | Every inquiry differs and a fast, specific answer wins the sale |
What can go wrong with an AI agent, and how do you control it?
An AI agent can give a wrong answer when the information behind it is thin, old or missing. That is the main risk, so the setup matters more than the technology. Five controls keep it in line:
- One approved answer list, signed off by you. The agent is set up to give only what is on it.
- Handover rules: custom quotes, complaints, refunds, and medical or legal questions go to a person.
- A way out: any customer can ask for a person at any time.
- Saying what it is: we recommend the first reply says it is your AI assistant.
- A monthly review of the chats, with the answers updated.
What do Meta’s WhatsApp rules say about bots and AI?
Automation needs an exit. The WhatsApp Business Messaging Policy (last updated 23 September 2026) allows automation inside the 24-hour customer service window but requires “prompt, clear, and direct escalation paths”: a human agent in the chat, a phone number, an email, web support, an in-store visit or a support form. The policy says automation, not AI, so it covers menu bots too.
AI providers cannot offer AI itself as the product. Section 4.7 of Meta’s Terms for WhatsApp Business Platform (last modified 23 September 2026) bars providers of AI technologies, such as large language models and general-purpose AI assistants, from using the platform to offer that technology when it is the primary functionality. Meta’s AI-providers page (updated 15 October 2026) adds that general-purpose AI assistants are allowed only where Meta is legally required to permit them: Brazil, Italy, Türkiye and a group of European countries. The UAE is not on the list.
Our reading, not Meta’s wording: an agent that answers for your own business is a different case, because the customer is asking about a class, not buying an AI assistant. Meta decides what counts as primary, so ask any provider how its setup fits. This is not legal advice.
Your chats cannot train anyone else’s AI model. If you keep an AI provider as your Solution Provider, section 4.7 says you may not let WhatsApp Business Platform Data be used to create, develop, train or improve AI models, except to fine-tune one for your exclusive use.
How do the 24-hour window and the handover to a person work?
Meta’s developer documentation (Service messages, updated 21 May 2026) sets the window: when a customer messages or calls your number, a 24-hour customer service window starts, and each new message or call resets it. While it is open, you can send free-form messages. After it closes, only pre-approved templates.
Menu bots and AI agents live under the same rule: after the window closes, even an AI agent can send only a template, so a quote left unanswered for 2 days needs one. We write your templates in Arabic and English and submit them to Meta, and the CRM sends the follow-ups on day 1, 3 and 7.
We set the handover triggers with you: custom quotes, complaints, refunds, health or legal questions, a customer who asks for a person, or an answer that is not on the list. The chat goes to the right person with the full history, and the agent says when to expect the reply, such as “tomorrow from 9 am”.
Where more than one system answers on a number, Meta’s Conversation Routing pages (updated 12 October 2026) give each conversation one owner at a time, who can pass control on. Passing control needs the business to accept Meta’s Business Agents terms.
Which one fits your business?
A menu chatbot is enough when customers ask a handful of stable things, nobody needs qualifying, booking or payment in the chat, and a person is on the number all day.
You need an AI sales agent when:
- every inquiry differs: the area, the date, the budget, the child’s age;
- customers write in dialect, or mix Arabic and English;
- ads bring inquiries at night and at weekends, and the first useful answer wins the booking;
- you want the buyer’s answers in the CRM before a person joins.
The agent is one piece of a connected system: it answers on the official WhatsApp Business API number, and the WhatsApp CRM follows up after it hands over. Training centers use the same pattern for parents’ questions, as in WhatsApp and CRM for training centers.