AI Chatbots on WhatsApp for Small Businesses: Where to Start

The free WhatsApp Business app on a phone offers quick replies, greeting messages, away messages, labels and a product catalogue. It is useful, but it cannot run a real AI chatbot because there is no supported way to connect an external system to it. Automation tools that claim otherwise usually rely on unofficial methods that risk the number being banned.

The WhatsApp Business Platform, usually called the API, is what chatbots run on. Meta offers it through its Cloud API, and most small businesses reach it through a Business Solution Provider such as Twilio, 360dialog, Wati, Respond.io or similar platforms. These providers handle number registration, give you an inbox for staff and connect to AI models or built-in bot builders.

What the API changes in practice

• Messages can be answered by software, with staff stepping in from a shared inbox.
• Outbound messages outside the 24-hour service window require pre-approved templates.
• Pricing is per conversation or per message category, set by Meta and passed on by the provider, so check current rates at the time of signing.
• Your business needs a verified Meta Business account and a phone number not already active on the regular app.

Start where questions are frequent, answers are stable and mistakes are cheap. A bot that tells someone your Saturday hours is low risk. A bot that quotes custom prices for a renovation job is not. Choose one or two of these for the first month:

• Opening hours, location and parking directions, with a map link.
• Booking and rescheduling, connected to a calendar such as Google Calendar or a booking system.
• Order status, pulled from a store platform like Shopify or WooCommerce.
• Menu, price list or catalogue questions, answered from a document you maintain.
• Lead qualification: collecting name, need, budget range and preferred time before a person calls back.

Modern AI bots answer from a knowledge source rather than rigid menus. You upload your price list, FAQs and policies, and the model writes the reply in natural language. That makes the quality of your source document the main factor in how good the bot sounds.

Limit what it may talk about. Write instructions that tell the model to answer only from the provided documents and to say it will pass the question to a team member when the answer is not there. Test it with awkward questions before launch: discounts you do not offer, medical or legal advice, complaints, and messages in a second language.

Keep the source short and current. A single page of verified facts outperforms a long folder of old brochures, because the model has less contradictory material to choose from. When prices change, update the source first and the bot follows.

Human handoff rules

Define triggers that move a chat to a person: the customer types words such as complaint, refund or speak to someone, the bot fails to answer twice, or the request involves money above a set amount. Tell the customer what happens next and when. A handoff that leaves someone waiting overnight without notice does more damage than having no bot.

WhatsApp requires opt-in before you send business-initiated messages, and its Business and Commerce policies restrict certain product categories. Data protection law applies too: in Turkey, KVKK governs personal data collected through chats, and GDPR applies to EU customers. Tell people a bot is answering, explain what you store and keep chat histories only as long as needed.

Avoid broadcasting promotions to everyone who ever messaged you. Users can block or report numbers, and a falling quality rating can limit how many people you are allowed to message.