🤖 Introducing Safety Link Preview Inside the qz-l Chatbot

Smart, automatic, AI-powered link safety — built right into your conversations.

We’re excited to announce a major enhancement to the qz-l.com AI chatbot:
our brand-new Safety Link Preview Tool, now embedded directly into your chat experience.

This upgrade makes the chatbot not just conversational, but actively protective whenever you interact with URLs.

🛡️ What Is the Safety Link Preview Tool?

Whenever you send a message containing a URL — or even phrases like:

“Check this link”
“Preview this URL”
“Is this site safe?”
“Show me what’s inside this webpage”

…the chatbot now automatically:

Detects your intent using our new built-in intent classifier
Extracts the link
Fetches content through our headless Safety Renderer
Collects metadata, OG tags, and visible text
Analyzes the page with Gemini AI using our safety prompt
Returns a rich, structured safety preview card inside the chat

The preview card includes:

📝 Summary
🔒 Safety classification
🏷️ Category
⚠️ Risk notes
🔗 Original link

Fast, clean, privacy-friendly — exactly what qz-l is built for.

🧠 How It Works Under the Hood

1. Intent Classification

Our enhanced system prompt guides the AI to automatically recognize when the user wants to:

preview a link
check safety
inspect a website
understand what’s behind a URL

This means you don’t need commands — the chatbot simply understands you.

2. Headless Safety Renderer (Playwright)

When a URL is detected, our internal render service:

Uses Playwright running on a tiny EC2 instance (2 vCPU, 1GB RAM)
Loads the page until DOMContentLoaded
Extracts:
- <title>
- <meta description>
- OG image, OG title, OG description
- Favicon
- Visible body text

We run this through:

A shared browser instance
A serial request queue
A memory-safe sandbox

Efficient, fast, and cost-optimized.

3. Gemini AI Safety Analysis

The extracted page data is passed to Gemini with our upgraded safety prompt, generating structured output:

{
  "summary": "...",
  "safety": "safe | maybe-unsafe | unsafe",
  "category": "...",
  "riskNotes": "..."
}

We improved the prompt to not only classify pages, but also advertise this feature naturally inside chat when appropriate.

4. Returning Structured Tool Results to the Chat UI

Our chat API now sends structured tool results:

{
  "role": "assistant",
  "content": "...",
  "toolResults": [
    {
      "type": "safety-preview",
      "success": true,
      "data": { ...analysis }
    }
  ]
}

This lets the frontend decide which React component to render (e.g., <PreviewLink />).
The "type" field makes the architecture extensible for future tools.

💬 Example Chat Flow

You:

Is this website safe? https://example.com

Chatbot:
(auto invokes the safety preview tool)

🔒 Safety Link Preview
Summary: Example Domain provides documentation demo content.
Safety: safe
Category: education
Risk Notes: No risky behavior detected.

All inside one friendly chat message — fast and intuitive.

🚀 Why We Built This

Short links are powerful, but they can also hide dangers.
qz-l.com is committed to being a:

privacy-first

safety-first

transparency-first

…URL platform.

The new chatbot safety preview helps protect users by letting them inspect links before clicking them — whether they’re shortening URLs, checking unknown links, or helping others stay safe online.

🔮 What’s Coming Next

With the new tool pipeline in place, we are exploring future additions:

🖼️ Screenshot previews
🧪 Malware-pattern detection
🔍 SEO / metadata reports
📦 Performance metrics
📝 Full text summarization

The architecture now makes these simple to plug in.

Try It Yourself

Visit https://qz-l.com
Open the chatbot
Paste any link
Get an instant safety preview — powered by qz-l’s own headless renderer + Gemini AI.

Introducing Safety Link Preview Inside the qz-l Chatbot