Proprietary · Statistical · 12 Metrics

Multi-Metric AI Scanner™

12-Signal AI Text Detection Engine

Multi-Metric AI Scanner™ is the core detection engine of TextHumanize. It computes 12 independent linguistic signals — from entropy distribution and n-gram frequency to stylometric fingerprinting — and aggregates them into a precise AI probability score calibrated via the Sigmoidal Calibrator™.

How It Works

1

Entropy Analysis

Computes character-level, word-level, and bigram Shannon entropy. AI text has suspiciously uniform entropy distribution; human text shows natural peaks and valleys across the document.

2

12-Metric Aggregation

Independently scores: text entropy, burstiness, vocabulary diversity (TTR, MATTR, Yule's K, hapax ratio), Zipf adherence, stylometry (function words, punctuation), AI patterns (overused connectors, phrase repetition), readability consistency, sentence opening diversity, and punctuation rhythm.

3

Weighted Voting

Each metric is weighted by its historical discriminative power on a 50k holdout corpus. Metrics with high variance between human and AI text receive higher weight in the final probability calculation.

4

Per-Metric Report

Outputs a per-metric breakdown alongside the aggregate AI probability score (0–100%), highlighting which specific signals flagged the text as AI-generated.

Key Metrics

12
Independent metrics
91.4%
Accuracy (holdout)
< 6%
False positive rate
50 words
Min text length
All (statistical)
Languages
< 30ms
Processing time

Use Cases

  • Auditing AI-generated content before publishing
  • Quality assurance in editorial and content workflows
  • Educational integrity checking for institutions
  • Baseline scoring for humanization pipeline feedback loops

Try Multi-Metric AI Scanner™ Now

Powered by our proprietary technology stack. Available directly in the TextHumanize tools and via REST API.

Related Technologies

Processing...