How It Works
What IS IT VIBE? does
IS IT VIBE? analyses public websites for fingerprints that suggest AI-generated scaffolding. Rather than guesswork, it scans live HTML, CSS, and JavaScript for patterns that rarely survive human editorial review once engineers or designers take over.
The tool produces a score from 0 to 100, with higher values indicating stronger evidence of AI involvement. Alongside the score, you receive a confidence assessment based on pattern diversity and severity.
The analysis process
- Fetch: The tool retrieves your site's HTML, CSS, and JavaScript. For single-page applications, it uses a headless browser to capture the rendered output.
- Detect: Over 40 pattern detectors scan for structural tells (div nesting, utility class spam), content tells (placeholder text, conversational comments), and stylistic tells (stock emoji sets, buzzword density, repeated gradients).
- Score: Each detection carries a severity level (critical, high, medium, low). The scoring algorithm weights these severities, applies diversity multipliers, and scales down for large enterprise sites to account for legitimate complexity.
- Validate: An optional LLM validator reviews the results and flags potential false positives or missed patterns, helping calibrate future detector improvements.
What the scores mean
- 0–19: Minimal AI involvement. Confidence depends on pattern diversity.
- 20–39: Some AI assistance detected, typically stylistic polish from tools like Copilot.
- 40–59: Noticeable AI patterns worth a deeper editorial pass.
- 60–79: Strong AI signature. Expect placeholder copy or duplicated structures.
- 80–100: Almost certainly prompt-and-paste. Critical detectors usually light up here.
Key detection categories
Critical signals
Conversational HTML comments ("Here's the navigation"), placeholder text (Lorem ipsum, TODO markers), and direct LLM signatures (ChatGPT, Claude, v0.dev mentions).
High-severity signals
Console logging in production, empty event handlers, educational comments, repeated DOM fragments, the "AI emoji toolkit" (🚀💡⚡✨), and suspiciously round social proof numbers (10,000+ users, 99% satisfaction).
Medium-severity context
Generic class names (container, wrapper), boilerplate framework text, generic testimonials, missing favicons, and systematic CSS variable scales typical of AI design systems.
Low-severity texture
Deep div nesting, stock photo sources, excessive font imports, and limited colour palettes.
Use cases
- Pre-launch review: Run the CLI against landing pages before sign-off to identify sections needing editorial attention.
- Agency deliverables: Point non-technical stakeholders at the web UI to assess agency submissions.
- Continuous monitoring: Wire the API into internal dashboards to catch regressions when content teams update CMS entries.
Dive deeper
The complete documentation covers architecture, configuration, testing, and calibration:
- Overview – explains the architecture and confidence model
- Getting started – covers installation, CLI flags, and API usage
- Detection catalogue – lists every pattern and severity tier
- Configuration – shows how to override detector thresholds
- Testing and calibration – documents the calibration dataset and batch runner