Writer tested 11 AI detectors — Claude-assisted fiction scored 100% human
A science fiction and fantasy writer uses Claude Code purely as an editorial tool for tracking character details and story continuity, not for coding. As publishing and debates intensify over whether writing counts as "real" writing, the writer ran an experiment: 11 different s were run against 8 pieces of provably human-written text, half of which predate consumer LLMs (written before 2019).
The results were counterintuitive. The piece scoring lowest on "humanness" was one written entirely by hand back in 2019, before any AI tools existed.
The piece scoring highest — rated 100% human by multiple detectors — was written this summer with Claude Code helping track characters and continuity. The writer argues that creative energy is a scarce resource best spent on actual sentences, not on manually updating notes and bibliographies, and sees no reason to avoid AI assistance for that work.
Key points
- Writer uses Claude Code as a continuity/character-tracking editorial assistant for fiction, not for coding
- Ran 11 s against 8 human-written texts, half predating 2019 (pre-LLM era)
- The 2019, fully hand-written piece scored lowest on "humanness"
- This summer's Claude-assisted piece scored 100% human on multiple detectors
- Argues s are unreliable and pushes back on publishing-industry AI bans