AnalysisPolicyJuly 15, 2026

AI detectors rate human writing as 5% human, Claude-assisted as 100% human

A writer using Claude Code found that AI detectors rated their pre-LLM writing as only 5% human, while Claude-assisted pieces scored 100% human. The author argues this shows AI detection is unreliable.

1 source

I use Claude Code as my continuity department for fiction I write myself. According to the publishing industry, that means I'm not a real writer. I tested 11 AI detectors on my pre-LLM writing and apparently, I'm as little as 5% human. My Claude-assisted pieces from this summer rate 100% human.originalseparation.substack.com

Back to the feed