In Turkey, when a customer wants to reach you, the first app they open is most likely WhatsApp. So how do you turn that channel from something that depends on a person typing in the green app on their phone into a 24/7, AI-powered customer support system that you run from your own interface? The answer: the WhatsApp Business Cloud API plus the Claude (Anthropic) API behind it.

This article doesn't skim the surface. We explain what the "new API" actually is, why you can't use that number on your phone, how to build your own WhatsApp interface with webhooks, and — the most misunderstood topic — message templates and the 24-hour rule from start to finish. Then we show the architecture of placing Claude behind it to run an autonomous support assistant.

What exactly is the "new API"? The WhatsApp Business Cloud API

The first source of confusion is naming. There are three different things:

  • WhatsApp (personal app): The regular WhatsApp on your phone.
  • WhatsApp Business App: For small businesses, still managed manually from a phone.
  • WhatsApp Business Platform (Cloud API): The programmatic API hosted in Meta's own cloud. This is what "the new API" means. It replaced the old On-Premises API you had to host yourself; now Meta runs the infrastructure and you simply connect to the API.

The Cloud API lets software (your system) send and receive messages over WhatsApp. In other words, it's not a human but code that talks. That's where both its power and its rules come from.

The critical point: you don't use this number on your phone

The most common misconception: "once I connect the API, I'll also check it from WhatsApp on my phone." No. When you register a phone number with the Cloud API, that number can no longer be used at the same time in the regular WhatsApp or WhatsApp Business app. It becomes an "API number."

So in practice we recommend using a separate number dedicated solely to the API for customer communication. If you also want to keep using your existing business number manually, don't migrate it to the API; set aside a new line. That way the system doesn't break because of the "let me also check it on the phone" expectation.

So where will you see the messages? The answer is in the next section: in your own interface.

Building your own WhatsApp interface with webhooks

The Cloud API works in two directions:

  • Inbound messages: When a customer writes to you, Meta sends the event to your server instantly as an HTTP POST via a webhook. You don't "poll" for messages; Meta pushes them to you. Delivery/read receipts arrive on the same channel.
  • Outbound messages: To reply, you make a request to the /messages endpoint of Meta's Graph API (with your phone number ID + access token).

These two ends let you build your own WhatsApp interface: capture every inbound message in the webhook, store it in your database (and your CRM), and show it as an "inbox" in your own panel. Agents reply from there, and the conversation history accumulates on the customer record. No green app needed. Verifying the webhook (verify token) and validating incoming requests with a signature (app secret) are the basic security steps.

Templates and the 24-hour rule — the most misunderstood topic

This is where everyone gets stuck, so let's be precise. To stop businesses from spamming people, WhatsApp enforces a strict rule:

The 24-hour customer service window

The moment a customer messages you, a 24-hour window opens. Within it you can send free-form messages (any content you like) — like a normal chat. Every new message from the customer extends the window back to 24 hours.

When the window closes, or never opened: only approved templates

Now the critical sentence: You cannot send a free-form message via the API to someone who has never written to you. Likewise, once the 24-hour window has closed, you can't send free-form text. In both cases the only way is to send a message template pre-approved by Meta.

So in a "I have 5,000 numbers, let me blast a campaign to all of them" scenario, if those people haven't messaged you before, you can only reach them with an approved template and their opt-in. Templates go through Meta approval and are split into categories:

  • Utility: Transactional messages like order status, appointment reminders, shipping info.
  • Marketing: Campaigns, announcements, re-engagement messages.
  • Authentication: One-time password (OTP) messages.

In short, the logic is: if the customer starts the conversation → you chat freely for 24 hours. If you want to start it → an approved template + opt-in is mandatory. Any system built without knowing this rule hits a wall on the very first campaign.

Note: Meta's conversation/template pricing is category-based and updated from time to time; confirm current pricing in Meta's official docs before you start.

Placing the Claude API behind it: an autonomous support program

Now the good part. You capture the inbound message in the webhook — but who writes the reply? This is where the Claude API comes in. The flow works like this:

  • 1. The customer writes on WhatsApp → Meta forwards the message to your server via webhook.
  • 2. Your server (e.g. Laravel) receives the message and gathers conversation history and customer context (from the CRM).
  • 3. You send this context, together with a system prompt, to Claude's Messages API: "You are the support assistant for brand X, follow these rules, you have this information…"
  • 4. Claude produces a reply. When needed, via tool use it reaches the real data in your system: look up an order, check stock, create an appointment, start a return.
  • 5. You send the reply back through the WhatsApp API as a free-form message if the 24-hour window is open. If it's closed, you continue with an appropriate template.

With this architecture your assistant stops being "a bot with canned answers" and becomes a reasoning support layer connected to your company's data. Thanks to Claude's tool use, it answers "where is my order?" with the real shipping status and "is there an appointment tomorrow?" with the real calendar. With prompt caching you cache the repeated system prompt to cut costs; with human handoff rules you route sensitive topics to a live agent.

What can you build with it?

  • 24/7 first-line support: Instantly answers FAQs, order/shipping tracking and the returns process, even at midnight.
  • Appointments and reservations: Clinic, salon, service center — the customer books via WhatsApp and the system writes it to the calendar.
  • Lead qualification: The assistant matures the incoming request with questions and routes qualified ones to sales.
  • Multilingual support: Turkish, English, German — Claude switches languages within the same assistant.
  • Order notifications: Transactional messages like "your order has shipped" (via utility templates).

Things to watch out for

  • Opt-in: You must have consent from the people you'll template-message; otherwise your number's quality rating drops.
  • Quality rating and messaging limits: Meta assigns a quality rating to your number and tiers the number of daily messages you can send. Spammy behavior lowers the limit.
  • Data protection (KVKK/GDPR): Set up explicit consent, retention and deletion correctly when processing customer data.
  • AI guardrails: Tell the assistant clearly what it can and cannot do; restrict things like price commitments or legal advice and hand off to a human.

Setup with Partnerfy

This system has four legs, and all four are our expertise: multi-channel messaging infrastructure, API integration (webhook + Graph API + template management), AI and automation (Claude integration, tool use, guardrails), and the CRM/ERP connection. From migrating the number to the API, to your own inbox interface and the Claude-powered assistant, we build it end to end.

If you want to automate your WhatsApp support line with AI, get in touch; we'll map out your scenarios and propose the right architecture for you.

Sources