DebotifyText NLP Research Lab
Empirical AI Detection
Research & Benchmarks
We publish independent empirical testing, token perplexity datasets, and structural burstiness analyses to establish transparency across automated AI detection models.
Published Studies & Whitepapers
Annual BenchmarkPublished October 2026 • 12 Min Read
The 2026 AI Detector Efficacy & Bypass Rate Report
An empirical evaluation of 10,000 synthetic, hybrid, and humanized text samples tested across Turnitin, GPTZero, Originality.ai, Copyleaks, and Winston AI. Includes false positive rates, entropy thresholds, and syntactic variance scoring.
Sample Size: 10,000 Documents5 Detectors EvaluatedTurnitin Bypass Rate: 99.9%
Read Full Paper Deterministic Perplexity
Analyzing word predictability across standard language models using n-gram cross-entropy and loss calculations.
Sentence Burstiness
Measuring natural standard deviations in sentence length, punctuation variety, and rhythmic cadence.
Semantic Anchoring
Preserving critical entities, technical citations, and argumentative intent during syntactic restructuring.