Stony Brook Study of 600 Amazon Books Finds AI-Flagged Titles Share Most Rare Phrases

2.7 · 4d ago · US · AI

The Facts

  • Tuhin Chakrabarty, a computer scientist at Stony Brook (State University of New York), studies detection of AI-generated text; a journalist cited his research in claiming significant portions of the 2026 bestselling novel Daggermouth were AI-produced. Nature only
  • The novel's author, H. M. Wolfe, has denied the allegation. Nature only
  • A July 2026 arXiv preprint by Chakrabarty and colleagues traced rare word sequences appearing verbatim in different books, to measure how much language books derive from earlier books. Nature only
  • The study compared three groups of Amazon books, 200 each: titles flagged by the AI-detection software Pangram as containing substantial AI-written text, self-published bestsellers with no AI-written text, and prize-winning or nominated literary fiction. Rare expressions were matched against a sample of hundreds of thousands of titles on Google Books. Nature only
  • Literary fiction had the least overlap with existing books, self-published bestsellers more, and AI-flagged books the most. Nature only
Dive deeper · 2 more factsShow less
  • In unpublished follow-up work, the researchers found AI-generated books also copy specific narrative beats from earlier human-written books. Nature only
  • The study did not name authors. Chakrabarty shared the underlying data with The Atlantic, whose reporter used it in the Daggermouth story. Nature only

Context

In an interview with Nature, Stony Brook computer scientist Tuhin Chakrabarty described research on detecting AI-written fiction by measuring verbatim overlap of rare phrases with earlier books. His work was cited in a journalist's allegation, denied by the author, that the novel Daggermouth was substantially AI-written.

Chakrabarty said short overlaps can occur by coincidence, especially in formulaic genre fiction, but longer and more specific matches are harder to explain that way. He also advocated disclosure of AI use. The excerpt was truncated, so further details are unavailable.

How the Agent Panel Scored This

Independent AI personas each scored importance 1–10; the story's score is a trimmed mean (highest and lowest votes dropped). Vote spread: 1.

3 Business Analyst This is a niche academic preprint about detecting AI-written fiction via phrase overlap, with limited immediate market impact beyond the publishing and AI-detection niche, and its concrete event is a research finding rather than a market-moving or regulatory action.
3 News Editor This is a niche academic preprint about detecting AI-written fiction via phrase overlap, with limited immediate impact beyond publishing and literary circles, though it touches a topical controversy.
2 Informed Citizen This is a preprint study of book text overlap with limited direct impact on ordinary readers, and its main concrete event is a research finding tied to a single contested allegation rather than a broad, immediate change.

Source

← All stories