DEV Community

אחיה כהן
אחיה כהן

Posted on • Edited on

Building an AI-Powered WhatsApp Bot with n8n and the Official Cloud API (2026 Guide)

#whatsappn8n #automation #tutorial #ai

Most WhatsApp bot tutorials are outdated or use unofficial APIs that get your number banned. Here's how to build a production-ready AI WhatsApp bot using the official Cloud API and n8n — the same architecture I use for all my client deployments.

Why This Stack?

Component What Cost
WhatsApp Cloud API Official Meta API for sending/receiving messages Free tier available (check Meta for current pricing)
n8n Open-source workflow automation (self-hosted) Free
AI Model (GPT/Claude) Natural language understanding + response generation ~$5-20/month
VPS Server Runs n8n + webhook endpoint $5-10/month

Total cost: ~$10-30/month for a bot that handles thousands of conversations. Compare that to SaaS platforms charging $50-500/month.

Architecture Overview 

Customer sends WhatsApp message
        ↓
Meta Cloud API receives it
        ↓
Webhook sends to your n8n instance
        ↓
n8n workflow processes the message:
  → Checks message type (text/image/location)
  → Loads conversation history
  → Sends to AI model with system prompt
  → AI generates contextual response
        ↓
n8n sends response back via WhatsApp API
        ↓
Customer receives reply instantly
Enter fullscreen mode Exit fullscreen mode

Step 1: Set Up WhatsApp Cloud API

  1. Go to Meta for Developers and create an app
  2. Add the "WhatsApp" product to your app
  3. Get your Phone Number ID and Access Token from the WhatsApp dashboard
  4. You get a test phone number immediately — no need to wait for approval

Important: Generate a permanent access token (System User token), not the temporary one that expires in 24 hours.

Step 2: Set Up n8n

If you don't have n8n running yet, the fastest way:

# Docker one-liner
docker run -d --name n8n \
  -p 5678:5678 \
  -v n8n_data:/home/node/.n8n \
  n8nio/n8n
Enter fullscreen mode Exit fullscreen mode

For production, I recommend Docker Compose with PostgreSQL (not SQLite) and Queue Mode for reliability. But for getting started, the above works fine.

Step 3: Create the Webhook

In n8n, create a new workflow:

  1. Add a Webhook node — Method: POST, Path: /whatsapp-webhook
  2. Copy the webhook URL (you'll need it for Meta)
  3. Add a Respond to Webhook node after your processing logic — Response Code: 200. This is critical — Meta requires a 200 response within 5 seconds or it retries.

Register the webhook with Meta:

  • Go to your app's WhatsApp Configuration
  • Add the callback URL: https://your-server.com/webhook/whatsapp-webhook
  • Subscribe to: messages

Step 4: Parse Incoming Messages

The webhook payload from Meta looks like this:

{
  "entry": [{
    "changes": [{
      "value": {
        "messages": [{
          "from": "972501234567",
          "type": "text",
          "text": { "body": "How much does shipping cost to Tel Aviv?" }
        }]
      }
    }]
  }]
}
Enter fullscreen mode Exit fullscreen mode

In n8n, use a Code node to extract:

const message = $input.first().json;
const entry = message.entry?.[0];
const change = entry?.changes?.[0];
const value = change?.value;

const msg = value?.messages?.[0];
const from = msg?.from;
const type = msg?.type;
const body = type === 'text' ? msg?.text?.body : '';
const name = value?.contacts?.[0]?.profile?.name || 'Unknown';

return { from, type, body, name };
Enter fullscreen mode Exit fullscreen mode

Step 5: Add AI Processing

Add an HTTP Request node to call your AI model:

{
  "model": "claude-sonnet-4-6",
  "max_tokens": 500,
  "system": "You are a helpful business assistant for [Your Company]. Be concise — WhatsApp messages should be short and clear.",
  "messages": [
    { "role": "user", "content": "{{$json.body}}" }
  ]
}
Enter fullscreen mode Exit fullscreen mode

Pro tip: Add conversation memory by storing previous messages in a Google Sheet or database, and including them in the messages array.

Step 6: Send the Response

Use another HTTP Request node to send the reply:

POST https://graph.facebook.com/v21.0/{PHONE_NUMBER_ID}/messages
Authorization: Bearer {ACCESS_TOKEN}

{
  "messaging_product": "whatsapp",
  "to": "{{$json.from}}",
  "type": "text",
  "text": { "body": "{{$json.aiResponse}}" }
}
Enter fullscreen mode Exit fullscreen mode

Making It Production-Ready

The basic flow above works, but here's what separates a demo from production:

Error Handling

If AI call fails → Send "Let me check on that, one moment..." → Queue for retry. If webhook times out → Return 200 immediately, process async.

Rate Limiting

WhatsApp has rate limits. Use n8n's built-in Wait node between messages if you're sending bulk notifications.

Media Handling

For voice messages: transcribe with Whisper. For images: describe with a vision model. Process the transcription with your AI.

Conversation Memory

Store conversations in PostgreSQL or Google Sheets. Key: phone number, Value: last N messages + timestamps. Clear after 24 hours of inactivity.

Human Handoff

When the AI can't handle a query: detect low-confidence responses, route to a human agent via Chatwoot or a simple notification.

Real-World Example

I built this exact architecture for a delivery company. The bot:

  • Receives shipment details from customers via WhatsApp
  • Calculates shipping prices based on distance (Google Maps API)
  • Documents everything automatically in Google Sheets
  • Sends status updates when deliveries are in progress

It runs 24/7 with high uptime. The company significantly reduced manual workload in customer service.

Common Mistakes to Avoid

  1. Using unofficial APIs — Baileys, green-api, etc. will get your number banned. The official Cloud API has a free tier — check Meta's current pricing for details.

  2. Not handling webhook verification — Meta sends a GET request to verify your webhook. Handle both GET (verification) and POST (messages).

  3. Slow webhook responses — Meta retries if you don't respond with 200 within ~5 seconds. Process messages asynchronously.

  4. No graceful degradation — When your AI provider has an outage, queue messages and retry.

  5. Ignoring message types — Customers send images, voice notes, and locations. Handle them or acknowledge them.

Resources

I'm Achiya, a WhatsApp automation specialist based in Israel. I've built 50+ WhatsApp bots for businesses using this exact stack. Questions? Drop them in the comments.

Top comments (0)