# AI assistants on WhatsApp Business > An AI assistant on WhatsApp runs on the WhatsApp Business Platform's Cloud API, not the WhatsApp Business app. It can reply freely within the 24-hour customer service window that opens when a customer writes; outside it, only approved template messages are allowed. Opt-in, a clear route to a human and a data transfer assessment under GDPR or KVKK are required parts of the setup. URL: https://corrplus.net/blog/whatsapp-business-ai-assistant/ Updated: 2026-10-27 Corrplus Team ## What is the difference between the WhatsApp Business app and the Platform? The WhatsApp Business app is what small businesses use to chat with customers by hand from a phone or computer. The WhatsApp Business Platform is the API layer for developers. In Meta's words, the Platform's Cloud API lets you message and call on WhatsApp programmatically. If you want to connect an AI assistant, a CRM or an order system to WhatsApp, this is the layer you use. For many businesses, in Türkiye and elsewhere, WhatsApp is the first channel a customer writes to. So the question is usually not whether to use WhatsApp, but at which layer and with how much automation. | Criterion | Business app | Platform (Cloud API) | Platform with an AI assistant | |---|---|---|---| | Multiple users | Limited | Through an inbox tool or CRM | Through an inbox tool or CRM | | Automation | Greeting and quick replies | Rule-based flows | Assistant that understands free text | | Template messages | No | Yes, with Meta approval | Yes, with Meta approval | | CRM and order system integration | No | Yes | Yes; the assistant can query data such as order status | | Human hand-off | A person is already replying | Defined in the flow | Defined in the flow, and required | | Fit | One person or a small team | Steady volume, several agents | High volume, repeated questions | Be clear about what the assistant is. A [chatbot](/glossary/chatbot/) is software that converses with users in text or voice; those built on language models understand free text but make mistakes if they are not bounded by rules. ## How do templates, the 24-hour window and opt-in work? Messaging on WhatsApp has two modes. When a customer writes to you, a 24-hour customer service window opens, and each new customer message resets it. Inside that window your assistant can reply without templates. The WhatsApp Business Messaging Policy states plainly that outside the window you may only send approved message templates. Templates are texts submitted to Meta and approved in advance. Each template is categorised as marketing, utility or authentication. Only templates with approved status can be sent, and a template can be paused or disabled based on quality feedback. Messages such as "your order has shipped", appointment reminders or payment confirmations go out this way. To message someone outside the window, you need their prior opt-in. Meta's rules are: state clearly that the person is opting in to receive messages from the business, state the business's name clearly, and comply with applicable law. Opt-in can be collected outside WhatsApp, for example through a web form, SMS, a phone menu or a paper form. Meta also recommends separate opt-ins per message category and requires honouring opt-outs. In practice, record where, when and with which wording you collected each opt-in. National rules on commercial electronic messages also apply; review them together with your legal counsel. ## When should the AI hand over to a person? The messaging policy allows automation within the 24-hour window, provided users get clear, direct ways to reach a person: in-chat transfer, phone, email or a support form. Hand-off is part of the rule, not an option. Write down the triggers in advance: - The user explicitly asks for a person. - A complaint, return request, payment issue or legal matter comes up. - The assistant fails to understand the same question twice, or finds no answer in its sources. - A sensitive situation involving health, safety or children is detected. Language models can produce answers that sound convincing but have no basis; this is called [AI hallucination](/glossary/ai-hallucination/). Restricting the assistant to approved sources (product catalogue, FAQs, policy texts) and having it query live systems for prices and stock reduces the risk. At hand-off, the agent should see the full conversation and the assistant's summary, so the customer never has to repeat themselves. Meta's Platform terms also restrict AI providers whose primary function is offering general-purpose AI assistants. An assistant that serves your own customers on specific topics is a different case. Still, keeping its scope tied to your business is the right call for both quality and the terms. ## What do GDPR and KVKK mean for a WhatsApp assistant? WhatsApp messages carry personal data: phone number, name and often order or booking details. Under GDPR and [KVKK](/glossary/kvkk/), Türkiye's data protection law, three questions stand out. The first is transparency: tell people what data is processed, for what purpose, and that they are talking to an AI assistant. The EU AI Act's transparency rules point the same way for users in the EU. The second is international transfer. Messages pass through Meta's infrastructure, and if the assistant uses a cloud language model, the conversation also goes to that provider. Both GDPR and Article 9 of KVKK (Law No. 6698) set conditions for such transfers, such as standard contractual clauses. The third is retention: decide how long you keep conversation logs and delete them when that period ends. Technical teams should not make this call alone. Map the data flow and decide together with your legal counsel. ## What drives the cost? We do not quote prices; rates change and vary by country. Three items drive the cost. - **Meta's message fees.** According to Meta's pricing page, charging is now per message rather than per conversation. The rate depends on the message category (marketing, utility, authentication, service) and the recipient's country code. Service messages inside the customer service window are free up to 1,000 per month per business phone number; since 1 October 2026, messages above that limit are charged. - **Model usage.** Every reply triggers the token cost of the language model behind it. Long histories and large knowledge sources raise it. - **Infrastructure and upkeep.** An inbox tool, CRM integration, keeping the knowledge source current and reviewing conversations regularly. ## Where should you start? 1. Pull the ten most frequent questions from last month's WhatsApp conversations. 2. Separate those an approved source can answer from those that need a person. 3. Write down the opt-in wording and record, the privacy notice and the hand-off rules. 4. Launch the assistant for a single topic first, such as order status or booking questions. We see this need often in [tourism and hospitality](/industries/tourism-hospitality/), where the same questions repeat in high season; in [ecommerce and retail](/industries/ecommerce-retail/), with heavy order and return traffic; and in [education](/industries/education/) during enrolment periods. For running a new customer's first steps on WhatsApp, see [customer onboarding automation](/solutions/customer-onboarding-automation/); for product recommendations, see [ecommerce personalisation](/solutions/ecommerce-personalization/). At Corrplus, discovery starts with conversation logs and knowledge sources. In design we set down the assistant's scope, hand-off rules and data flow in writing. We build in short cycles and, once live, review conversations on a regular schedule. More on our [AI chatbot development](/services/ai-chatbot-development/), [AI customer service](/services/ai-customer-service/) and [AI sales assistant](/services/ai-sales-assistant/) pages. ## Frequently asked questions ### Can we connect AI to the WhatsApp Business app? Not to the app itself: to connect your own AI assistant programmatically, you need the WhatsApp Business Platform's Cloud API. The app is designed for small volumes handled by hand on a phone. If you need automation, CRM integration or several agents, move to the Platform. ### Can we message customers whenever we like? No. You can send free-form messages during the 24-hour customer service window after the customer's last message. Outside it, only templates approved by Meta can be sent, and the person must have opted in beforehand. ### What if the AI gives wrong information? Language models can produce unsupported answers. Reduce the risk by restricting the assistant to approved content, setting firm rules for prices, returns and health topics, and handing the conversation to a person when it is not sure. ### Can we offer a general ChatGPT-style assistant on WhatsApp? Meta's Platform terms restrict AI providers whose primary function is offering general-purpose AI assistants, and Meta decides what counts as primary. An assistant that serves your own customers on orders, bookings or support is a different case. Keep the scope narrow and check the current terms. ## Sources - [Meta for Developers, WhatsApp Business Platform](https://developers.facebook.com/documentation/business-messaging/whatsapp/about-the-platform) - [Meta for Developers, Template fundamentals](https://developers.facebook.com/documentation/business-messaging/whatsapp/templates/overview) - [Meta for Developers, Getting opt-in](https://developers.facebook.com/documentation/business-messaging/whatsapp/getting-opt-in) - [Meta for Developers, WhatsApp Business Platform pricing](https://developers.facebook.com/documentation/business-messaging/whatsapp/pricing) - [WhatsApp Business Messaging Policy](https://whatsappbusiness.com/policy/) - [Meta Terms for WhatsApp Business Platform](https://www.facebook.com/legal/Meta-Terms-for-WhatsApp-Business-Platform) - [KVKK, Personal Data Protection Law No. 6698 (English)](https://www.kvkk.gov.tr/Icerik/6649/Personal-Data-Protection-Law)