Skip to content

November offer: 50% off your November fee on monthly plans, or off a full year on yearly plans.

View Plans

How to Set Up a WhatsApp AI Customer Agent in 5 Steps

An AI agent is only as good as the information and rules you give it. We work through it in order: which questions it takes on, which source it builds answers from, how it speaks and when it hands the chat to your team. At the end you'll find a pre-launch test checklist and the cost items.

Setting up a WhatsApp AI customer agent is less like switching an app on and more like hiring a new team member. You need to tell it what it knows, how to talk, which questions to answer and when to call you in. Think of it as writing a job ad. Below we walk you through it in five steps, along with the mistakes we see most often in the field.

What exactly does a WhatsApp AI customer agent do?

A classic chatbot follows the menu you draw: the customer types "1" and the bot shows the next step. An AI agent understands the customer's own words and builds its answer from the information you've given it. When a customer asks "How many days do I have to return something, and do you take items with the tag removed?", it doesn't show a menu. It just answers.

In Chatopya, the AI answers frequent questions 24/7 using the business's own information. Because it follows the conversation history, it understands what the customer means when they say "that product" in a second message, and it copes well with typos, abbreviations and idioms. Anything it can't resolve, it hands over to an agent on your team together with the chat history.

In short: an AI agent doesn't do every job. It takes on the repetitive questions, and the work that needs a decision stays with your team.

Type of workBest fitWhy
Open-ended frequent questions (returns, sizing, delivery, opening hours)AI agentCustomers phrase the question differently every time; it doesn't fit into a menu.
Step-by-step data collection (appointments, order forms, surveys)Rule-based flow built with the flow designerSteps follow a fixed order; you want the same information from every customer.
Complaints, exceptions, case-by-case decisionsHuman agentNeeds authority and empathy; a wrong answer is costly.

Meta's AI rule The WhatsApp Business Platform terms were updated on 15 January 2026 and restricted providers whose main service is offering an AI assistant. An agent that answers questions about your own products and services is a tool that supports your business; even so, it's safest to keep your agent limited to topics related to your business.

What do you need before you start?

  • A number connected to the WhatsApp Business API. The AI agent runs on the API side, not in the WhatsApp Business app on your phone. You can migrate your existing number; in that case the app stops working on that number and past chats aren't carried over. You can also start with a new number if you prefer.
  • Your information sources. Price list, return and exchange terms, delivery policy, product descriptions, FAQs.
  • At least one person to take handovers. And that person's working hours.
  • AI credits. AI Bot usage runs on credits; we explain the details in the cost section.

Step 1: Write down which questions it will answer

A good agent has a job description. Open your WhatsApp chats from the last month or two and list the 20-30 questions that come up most often. This list will be both your job description and, later on, your test scenario.

Questions it will answer

  • Delivery times, shipping costs, return and exchange terms
  • Product features, size and measurement details, usage tips
  • Opening hours, address, payment options
  • "Which one is right for me?" questions: the agent can ask about the customer's needs and recommend a suitable product from the catalog

Topics it won't handle

  • Personal discounts and haggling
  • Return or compensation decisions that fall outside the rules
  • Where a specific order is in transit: this lookup can be set up through a custom integration with your business system; if it isn't, hand the chat to an agent
  • Questions that need expert opinion, such as health or legal matters

Step 2: Define the knowledge source

An AI agent only knows what you give it. In Chatopya, you add your business information and your website pages to the knowledge base in the Bot Center, and the agent builds its answers from that source.

Bring scattered documents together

In most businesses, information is scattered. Prices sit in a PDF, the return policy in a Word file, product measurements in an Excel sheet. Gather everything in one place and clear out the contradictions. If two documents state two different return periods, your agent will be inconsistent too.

Fill the knowledge base step by step

  1. Pick what's current. Leave out finished campaigns, old prices and products you no longer sell.
  2. Split it into topics. Create sections such as delivery, returns, payment, products and opening hours.
  3. State rules clearly. Instead of "It depends", use plain sentences like "You can return items within 14 days of delivery."
  4. Add your website pages. If you already keep your FAQ, delivery and returns pages up to date, link them to the knowledge base as sources.
  5. Give every piece of information an owner. Make it clear who updates the knowledge base when prices or the shipping carrier change.

If you've connected your Trendyol, Hepsiburada (both major Turkish online marketplaces) or Shopify store, stock and price information is refreshed from the store at regular intervals, so you don't have to type it in by hand. We explain what belongs in the knowledge base in detail in our chatbot knowledge base guide.

It doesn't make up answers it doesn't know When a question comes in that has no match in the knowledge base, the agent doesn't guess; it hands the chat to your team. So an incomplete knowledge base leads to more handovers, not wrong answers.

Step 3: Set the tone

When customers chat with your AI agent, they're really chatting with your brand. Put your tone decisions into a short list:

  • Form of address: formal or informal?
  • Emoji: never, occasionally or often.
  • Answer length: on WhatsApp, answers of two or three sentences are easy to read.
  • Template for saying no: what will you say instead of "Unfortunately, we can't do that"?
  • Phrases to avoid: words that don't fit your brand, overly formal wording.

Reflect these decisions in the texts in your knowledge base too. If the source text is written in formal language, a warm tone is harder to strike. If the canned replies your human agents use are written in the same voice, customers won't notice a break at the moment of handover. If you have international customers, the agent can reply in several languages, Turkish first among them.

Step 4: Decide when to hand over to a human

The most valuable skill of a good AI agent is being able to say "I'm passing you to a colleague" when it's needed. Handover should kick in in these cases:

  • No answer in the knowledge base: the chat goes to an agent with its history, so the customer doesn't have to explain everything again.
  • The customer is angry or worried: sentiment analysis tags this customer instantly so your team can prioritise them.
  • A decision is needed: topics such as exceptional returns, damaged products or invoice corrections.
  • The customer explicitly asks for a person: don't push back, transfer them.
  • A banned word comes up: the words you define are detected and a notification goes to the manager.

Working hours and the handover rule

You set your working hours and handover rule in the Bot Center. Think ahead about what a customer handed over at 23:00 should be told: a clear expectation like "Our team will get back to you at 09:00 tomorrow" is better than a vague "as soon as possible". Because several agents can use the same number, you can assign a handed-over chat to the right person.

Account for the 24-hour window If more than 24 hours pass since the customer's last message, you can no longer send them free-form text; you need an approved template message. Get back to chats you handed over outside working hours before that window closes. Details are in our article on the WhatsApp 24-hour rule.

See it with your own questions. Let's try together in a demo how the AI agent answers using your information.

Book a free demo

Step 5: Test before going live

You can run pre-launch tests in the Bot Center. Use the question list from step one here, but ask the questions the way a real customer would:

  1. Ask the same question in three different ways.
  2. Type with typos and shorthand: "how mny days to return", "is shiping free".
  3. Push the context: first "Do you have the black one in L?", then "Ok, when would it arrive?"
  4. Ask something that isn't in the knowledge base. Does the agent hand over without making things up?
  5. Write an angry message and see how the handover works.

Spotted a wrong or incomplete answer? Fix it in the knowledge base, then try the same question again.

What should you monitor after going live?

The agent's quality is directly proportional to the quality of the knowledge base, so it needs regular upkeep in the first weeks. We recommend keeping an eye on three things:

  • Handed-over chats: Read the chats that land with your team once a week. If the same question keeps getting handed over, that information is missing from the knowledge base.
  • Satisfaction survey: The CSAT survey sent after every conversation shows how the answers land with customers.
  • Analytics dashboard: You track response time, resolution rate and satisfaction in one place.

What determines the cost of an AI customer agent?

There are two separate items. On the Chatopya side, AI Bot usage runs on credits: the Pro plan includes 500 AI credits and the Business plan 1,000, and you can buy extra credit packs if you need more.

Meta's side is separate. Since 1 October 2026, service messages sent within the 24-hour window after a customer writes are also charged; each number gets 1,000 free service messages a month, after which messages are billed individually. The answers your AI agent gives inside the window count as service messages too. Chatopya adds no commission to Meta's unit price. For current plan contents, see our pricing page; we also explain how the AI agent works in WhatsApp customer service on our WhatsApp customer service and AI chatbot page.

Frequently asked questions

What does a WhatsApp AI customer agent say when it doesn't know the answer?

It doesn't make up an answer to a question that has no match in the knowledge base. It hands the chat, along with its history, to an agent on your team.

Can I use an AI agent in the WhatsApp Business app on my phone?

The AI agent in Chatopya runs on the WhatsApp Business API. You can migrate your existing number to the API; in that case the app stops working on that number and past chats aren't carried over.

How long does setup take?

Setting up the panel takes about 5 minutes; if the number has never been registered with Meta before, Meta's own review time is added. The part that really takes time is preparing the knowledge base. Setup and training are free.

How often should I update the knowledge base?

Every time a price, campaign, delivery or return condition changes. Because the agent builds its answers from the knowledge base, outdated information reaches customers for as long as it stays there.

Share this article WhatsApp LinkedIn X Facebook
Related Chatopya solution Explore the chatbot solution

Let's set up your AI agent together

Prepare your knowledge base, handover rules and working hours together with our Turkish-speaking support team. Setup and training are free.

View Plans