Which AI wrote this?

Paste any text. We compare it against a corpus of known model outputs using gzip compression distance — no training, no external model call.

Longer text gives a more reliable answer.
How this works, and the full breakdown

Method

Normalized Compression Distance + k-nearest-neighbours (Lee et al., MobiSys ’24, based on Jiang et al.'s gzip text classifier). Two texts that gzip well together tend to share the same author's patterns — no model weights, no embeddings, just compression. Paper ↗

Reference corpus

13
ox-alpha
60
Gemini
60
GPT
60
Claude
53
GLM