Study Flags Daggermouth as 60% AI-Written as 20% of Kindle Ebooks Show Substantial AI Text
Updated
Updated · The Atlantic · Jul 27
Study Flags Daggermouth as 60% AI-Written as 20% of Kindle Ebooks Show Substantial AI Text
2 articles · Updated · The Atlantic · Jul 27
Summary
More than 14,000 Kindle ebooks screened with the Pangram detector found Daggermouth at 60% AI-written and one in five titles above the study’s 25% “substantial AI text” threshold.
Researchers led by Stony Brook’s Tuhin Chakrabarty said the novel also shared multiple rare five-word-plus phrases with other suspected AI books, a second check meant to reduce reliance on detector scores alone.
H. M. Wolfe denied using generative AI, calling the claim “wholly untrue,” while Simon & Schuster said it stands behind the book and noted it underwent the publisher’s normal editorial process.
The study is not yet peer-reviewed and does not prove AI use, but an outside expert said a fully human-written book scoring 60% AI would be “almost statistically impossible.”
Beyond Daggermouth, the study estimated AI-assisted books make up 20% of Amazon’s ebook catalog, 12% of sales and 10% of genre best sellers, pointing to a broader disclosure problem as AI-written fiction improves.
If AI detectors flagged Daggermouth as 60% machine-written, how should readers, Amazon, and publishers decide what counts as real authorship?
Did a Kindle romance hit a publishing jackpot with human talent alone—or is Daggermouth proof AI-assisted fiction can already become a bestseller?
As AI-assisted ebooks spread across Amazon, will disclosure rules arrive before more breakout novels blur the line between writer and machine?
AI Detection, Authorship, and the Crisis of Trust: How Automated Content and False Positives Are Reshaping Publishing in 2026
Overview
This report explores the growing challenges facing writers and publishers in the age of AI. It begins with the controversy over the Stony Brook University study, where selective data use led to exaggerated claims about fish population decline. The report then examines how AI detectors, trained on formal academic writing, often misclassify original human work—especially when writers use AI tools for editing or write in specialized styles—leading to false accusations. As AI-generated content floods platforms like Amazon, new royalty models and DRM policies further complicate author earnings and content control. These trends erode reader trust, threaten author livelihoods, and spark debates about the true meaning of authorship and literary value.