How to set up an AI agent for Arabic customer service on WhatsApp

The Jwab team · 19 October 2026 · 3 min read

An AI agent can answer most routine customer questions instantly, at 2 am, in the customer’s own dialect. But a badly set-up agent gives vague answers, invents things or traps customers who need a person. This guide shows how to set one up properly, using Jwab as the example. The same principles apply anywhere.

Step 1: Decide what the agent is for

Give each agent one clear job. “Answer everything” is how agents go wrong. Good examples:

Sales: prices, availability, delivery and how to order.
Support: how-to questions, order status, returns.
Bookings: services, available times, preparation, location.

In Jwab you can create several agents, each with its own name and department, so each job gets its own agent.

Step 2: Fill the knowledge base with real answers

The agent answers from your knowledge base, so its quality depends on what you put there. Write short, specific notes:

Prices, packages and what’s included.
Delivery areas, times and fees.
Exchange, return and refund policies.
Opening hours, branches and contact details.
Your 20 most frequent customer questions, with the answers your best employee would give.

Write them in plain Arabic (or Arabic and English). Specific beats general: “Delivery to Cairo and Giza within 24 hours, free over EGP 500” is far better than “we deliver fast”.

Step 3: Write clear instructions

Instructions tell the agent who it is and how to behave. A good structure:

Identity: “You are Laila from the sales team at [business].”
Tone and language: “Reply in a friendly Egyptian dialect, or in English if the customer writes in English. Keep replies short.”
What to do: “Help customers choose a product and explain how to order.”
What not to do: “Never promise discounts. Never give medical or legal advice. If the answer isn’t in the knowledge base, say you’ll check with the team.”
When to hand over: “If the customer asks for a person, complains, wants a discount or is ready to pay, hand the conversation to the team.”

Dialect matters in Arabic. Telling the agent which dialect to use, or to mirror the customer’s, makes replies feel natural instead of formal and stiff.

Step 4: Plan the handover

The best AI agents know when to stop. In Jwab, when an agent hands over, the conversation goes to the department and person the manager chose, and the team gets an alert. Make sure:

Someone in that department is actually available, or customers are told when to expect a reply.
Your out-of-hours message sets the right expectations.
The team reads the conversation (or its AI summary) before replying, so the customer never has to repeat themselves.

Step 5: Test before going live

Use the preview to ask the agent the questions customers really ask, including the awkward ones:

Questions in different dialects and in “Franco” Arabic (Arabic written in Latin letters).
Questions the knowledge base doesn’t cover. The agent should say so, not invent an answer.
Requests for a discount or a person. The agent should hand over.

Fix the knowledge base or the instructions after each wrong answer, then test again.

Step 6: Go live gradually and keep improving

Start with one channel or department, read a sample of the agent’s conversations every day for the first week, and add every new question to the knowledge base. Agents get better as your knowledge base grows.

Be transparent

Let customers know they’re talking to an AI agent, and always give them a way to reach a person. It builds trust, and on WhatsApp it is part of Meta’s rules for automated messaging.

AI agents and the knowledge base are part of Jwab’s Professional and Business plans. See the plans or ask for a 7-day free trial in our chat.