DEV Community

jesus manrique
jesus manrique

Posted on • Originally published at guayoyo.tech

You Approve, AI Publishes: Human Approval Circuit via WhatsApp — Part 4 of 5

#ai #automation #productivity #tutorial

Your pipeline already generates copy and images. The next step is the most important decision: publish or not?

The "human-in-the-loop" pattern is what separates an AI toy from a real production tool. The human decides. The AI executes. And the most immediate channel for that decision: WhatsApp.

The Approval Architecture 

AI generates copy + image → WhatsApp Preview (text + image)
                                ↓
                           You reply:
                      ┌──────┴──────┐
                     OK             NO (+ feedback)
                      ↓              ↓
                Publish to      Regenerate with
                IG/TikTok       your corrections
                      ↓              ↓
           WhatsApp: "✅       Resend new
           Published"           preview
Enter fullscreen mode Exit fullscreen mode

Total review time: 30 seconds. Open WhatsApp, see the image, read the copy. If you like it, "OK." If not, "NO, make it more technical." The AI understands natural language feedback.

WhatsApp Integration Options

Option A: Evolution API (recommended for self-hosted)

Evolution API is an open-source WhatsApp gateway with a clean REST API:

docker run -d -p 8080:8080 atendai/evolution-api

# Send text message
POST http://localhost:8080/message/sendText/evolution
Headers: apikey: YOUR_API_KEY
Body: {
  "number": "584140108660",
  "text": "📱 *New post ready for review:*\n\n{copy}\n\nOK or NO?"
}

# Send image
POST http://localhost:8080/message/sendMedia/evolution
Headers: apikey: YOUR_API_KEY
Body: {
  "number": "584140108660",
  "mediatype": "image",
  "caption": "Post preview",
  "media": "https://your-server/images/preview.png"
}
Enter fullscreen mode Exit fullscreen mode

Option B: Meta WhatsApp Cloud API (official)

Requires a verified Meta Business Account. More bureaucratic but more stable:

POST https://graph.facebook.com/v21.0/{phone_number_id}/messages
Headers: Authorization: Bearer {meta_token}
Body: {
  "messaging_product": "whatsapp",
  "to": "584140108660",
  "type": "image",
  "image": { "link": "https://your-server/images/preview.png", "caption": "OK or NO?" }
}
Enter fullscreen mode Exit fullscreen mode

Option C: Direct webhook to OpenClaw (used in this series)

If your AI assistant is already connected to WhatsApp via OpenClaw, the n8n webhook sends the preview and OpenClaw listens for your response:

POST https://ai.guayoyo.tech/plugins/webhooks/approval
Body: {
  "action": "send_preview",
  "recipient": "+584140108660",
  "copy": "post text...",
  "image_url": "https://...",
  "brief_id": "post-2026-05-14-001"
}
Enter fullscreen mode Exit fullscreen mode

The Circuit in n8n

Node 1: Send Preview via WhatsApp 

// HTTP Request → Evolution API
const preview = {
  number: "584140108660",
  text: `📱 *New post to review*\n\n${$json.hook}\n\n${$json.body}\n\n${$json.hashtags}\n\n${$json.cta}\n\n---\nReply *OK* to publish or *NO* + feedback`,
};
// If image exists, send image first then text
Enter fullscreen mode Exit fullscreen mode

Node 2: Wait for Response (Webhook)

n8n exposes a Webhook node that waits for the response:

Webhook Node (POST /approval-response)
  ↓
IF node: Does text contain "OK"?
  ├─ Yes → Continue to publishing
  └─ No → Extract feedback → Regenerate
Enter fullscreen mode Exit fullscreen mode

Node 3: Switch — OK or NO 

const response = $input.first().json.body.toLowerCase();

if (response.includes("ok") && !response.includes("no ok")) {
  return [{ action: "publish", data: $input.first().json }];
} else {
  const feedback = response.replace(/^no[,\s]*/i, "").trim();
  return [{ action: "regenerate", feedback: feedback, data: $input.first().json }];
}
Enter fullscreen mode Exit fullscreen mode

Node 4: 30-Minute Timeout

If you don't respond within 30 minutes, the flow saves a draft and notifies you:

const draft = {
  copy: $json.copy,
  image_url: $json.image_url,
  status: "draft_timeout",
  timestamp: new Date().toISOString()
};
// Save to database or Google Sheet
// Send WhatsApp: "⏰ The 10am post is in drafts. Review when you can."
Enter fullscreen mode Exit fullscreen mode

Complete Circuit Flow 

1. Agent generates copy + SD generates image
              ↓
2. WhatsApp Preview: "📱 New post for review"
              ↓
3. Webhook waits for your response (30 min timeout)
              ↓
         ┌────┴────┐
        OK         NO + feedback
         ↓          ↓
4a. Publish      4b. Regenerate
     IG/TikTok      (back to agent
     ↓              with feedback)
5. WhatsApp:        ↓
   "✅ Done"     New preview
Enter fullscreen mode Exit fullscreen mode

Why WhatsApp and Not a Dashboard

Web Dashboard WhatsApp
Requires browser + login Already open
You have to go find it It comes to you
~2-3 minutes per review ~30 seconds per review
Desktop only Anywhere

Friction kills consistency. If publishing requires 3 extra steps, by day three you stop doing it. With WhatsApp, it's a reflex: see, respond, done.

Security: Only You Can Approve

The approval webhook only accepts responses from your number:

if (incoming.from !== "+584140108660") {
  return { status: "ignored", reason: "not_authorized" };
}
Enter fullscreen mode Exit fullscreen mode

No one else can approve posts. The circuit is human, but the security is programmatic.

Ready for a content pipeline where you decide and AI executes? At Guayoyo Tech, we build the complete circuit — from generation to publishing, with your approval at every step.

Top comments (0)