Back

Documentation

How VibeDetect analysis works

This page explains how the vibe code detector combines analyzer evidence into a score. If you are using VibeDetect as an AI generated website checker, this is the reference for how weighting and verdict thresholds are applied.

Scoring process

  1. 1. Run website or repository analyzers and capture category scores with confidence.
  2. 2. Normalize analyzer weights and compute contribution points.
  3. 3. Cap heuristics alone at Some AI Signals unless several categories agree; a tool fingerprint sets a floor and names the tool. Agent commits count by their share of the history, and a configured AI tool alone is Some AI Signals.
  4. 4. Map final score to a verdict range.

Website analyzer weights

CategoryWeight
AI Copywriting20%
Template Detection16%
Modern AI16%
UX Depth12%
Security10%
AI Analysis12%
Naming Patterns7%
Dependencies7%

Repository analyzer weights

CategoryWeight
AI Config14%
Commit Patterns12%
Velocity10%
Code Quality10%
Evolution7%
Git Workflow7%
README Quality6%
Test Coverage6%
Security6%
File Structure6%
Dependencies6%
AI Analysis6%
Social Patterns4%

Verdict thresholds

up to 24

Likely Human-Built

up to 49

Some AI Signals

up to 74

Probably Vibe Coded

up to 100 (inclusive cap)

Almost Certainly Vibe Coded

FAQ

How should I read the final VibeDetect score?

Scores range from 0 to 100. Don't read into it too much! It's only meant to be a bit of fun and a slight guide. Lower scores indicate more human-like patterns, while higher scores indicate stronger AI-generated or vibe coding signals.

Does VibeDetect use only AI models to classify code?

No. Heuristic analyzers always run first. Optional LLM analyzers are additive and intended to supplement deterministic signals.

Can false positives happen in detect AI code workflows?

Yes. Repeated design systems, mature templates, or deliberate style choices can resemble AI-assisted output. Treat results as a review aid, not final proof.