Could you spot an AI-written book?
TL;DR
AI prose has tells, says writer Imogen West-Knights — negative parallelisms, overblown metaphors, dense adjective stacks and repetitive syntactic blocks. Even so, publishers struggle to reliably flag AI-written books, because the models are trained on both great and terrible human writing. The issue grows as lightly edited AI books are increasingly marketed as “human-written”. For readers, the line between human and machine prose keeps blurring.
Nauti's Take
Encouraging: there are now identifiable stylistic tells that give AI prose away — a useful skill for editors, libraries and critical readers. The problem: publishers struggle to reliably flag AI books, because models train on both great and terrible writing, and the market is filling with lightly edited "AI-light" titles marketed as human-written.
Anyone after authenticity should pick more deliberately and favor transparent publishers.
Sources