The Atlantic: Is this what comes after AI slop?
Will Oremus’s July 2026 piece in The Atlantic uses a single book as the lens for a broader question about AI-assisted writing: what happens when AI-generated content is polished enough to succeed commercially and opaque enough to frustrate detection?
The case
Daggermouth by H.M. Wolfe spent months on USA Today’s bestseller list and ranked first in Amazon’s science-fiction romance category. Researchers studying AI content in self-published ebooks flagged it as the most popular title in a dataset of books containing substantial AI text. Tuhin Chakrabarty, a researcher who analyzed the text, found that roughly 60 percent of the book was flagged as AI-assisted by detection tools, with identical phrases appearing across multiple self-published ebooks — a pattern consistent with shared AI generation pipelines rather than coincidental phrasing.
H.M. Wolfe denied the accusations. BookTok communities split between readers defending her based on their reading experience and others pointing to the statistical evidence.
What the article adds
Oremus does not treat the Daggermouth case as isolated. He situates it within a broader finding: approximately one-fifth of Amazon ebooks now contain substantial AI content, based on the same research. Detection tools remain unreliable — current methods produce both false positives and false negatives at rates that make definitive attribution legally and practically difficult. The taboo against AI authorship in publishing is real, Oremus notes, but it is difficult to enforce without reliable detection.
The article also addresses the economics. Authors who use AI assistance face strong incentives to conceal it: reader backlash is steep, publishing deals that have been announced with AI-generated manuscripts have faced cancellations and contract disputes. Yet the commercial payoff can be significant when the output is compelling enough to attract organic readership. Oremus describes this as a structurally unstable situation — the gap between the incentive to use AI and the incentive to hide it will not resolve through individual author choices.
For writing practitioners
The article is not a how-to guide, but it surfaces several questions relevant to writers, editors, and content teams:
On detection. Current AI detection tools are imperfect enough that they cannot serve as a reliable editorial gate. Teams relying on detection software to enforce AI disclosure policies should understand what the error rates mean in practice.
On disclosure norms. The publishing industry has not converged on where to draw the line between AI-assisted editing, AI-assisted drafting, and AI-generated text. The Daggermouth case illustrates the cost of that ambiguity for authors, readers, and publishers alike.
On market signals. The book’s commercial success before the controversy suggests that AI-assisted fiction can achieve reader engagement without readers identifying it as AI-generated. Whether this finding changes how content teams think about AI assistance depends on how they weigh audience reception against disclosure ethics.
Oremus ends with a forecast rather than a conclusion: the tension between AI-assisted production and commercial disclosure norms will eventually require industry-level policy rather than individual author policing. The article is worth reading for anyone working in content-heavy environments where AI use is common but undeclared norms still govern what gets published under whose name.