Hachette's imprint Orbit recently pulled the horror novel Shy Girl by Mia Ballard from its planned US and UK release after The New York Times brought AI-generation allegations to the publisher, originating from a mid-2025 Reddit post that flagged repetitive patterns often associated with AI-generated prose: nearly every noun paired with an adjective, actions described through similes, descriptions arriving in threes. Concerns spread across BookTok, Instagram, and YouTube before The Times investigated. Ballard denies personally using AI, saying an acquaintance she hired to work on the book's original, self-published version incorporated AI tools without her full knowledge. Hachette has said only that it conducted a "thorough and lengthy review" and remains committed to "protecting original creative expression." No conclusive evidence has been made public either way.
I want to sit with a question underneath this story that I don't think gets asked enough: what's actually the difference between an author using AI as an editing tool because a professional edit isn't financially within reach, and an author having AI generate the book itself? Right now, the industry treats those as the same red flag, but they aren't the same act.
The Line That Actually Exists
A real, defensible distinction exists, and it's the same one the US Copyright Office draws. AI-generated expression on its own isn't copyrightable, but a human's creative choices are, even when AI assists at some point in the process. A tool that flags pacing problems, overused words, or structural issues doesn't write a single sentence of the book. The author still makes every word choice. That's assistive use. A tool that generates whole paragraphs, scenes, or plot beats means the author didn't write the sentences on the page, even if the author directed the prompt. That's substitutive use. One keeps a human as the author of the expression. The other doesn't. From the outside, both can produce prose with the same statistical fingerprints, repetitive sentence structures, certain overused words, which is exactly why detection tools and Reddit-level scrutiny can't reliably tell them apart. But the acts themselves aren't the same just because they can look similar on the page.
Why the Industry Reacts the Way It Does
Once a manuscript's authorship becomes publicly contested, a publisher cares less about proving what happened and more about limiting its own exposure. Pulling a deal before release, even without conclusive proof, is the lower-risk move: it avoids putting out a book under active public suspicion and avoids the deeper legal question of whether the publisher can even secure clean, enforceable rights over content that may not be fully copyrightable. That's a business decision made under uncertainty, not a verdict on what the author actually did.
Where This Lands Hardest
Here's the part that doesn't get said enough. The entire framework around "undisclosed AI use" as a red flag assumes the real alternative is always a human professional the author could otherwise afford. For most indie authors, that assumption is false. A full developmental edit on an 80,000-word novel runs around $2,400. Line editing adds roughly $3,200 more. A professionally narrated audiobook runs $100-$250 per finished hour, putting a standard novel at $600-$1,500 or more, before any mastering. Audible has become the dominant place readers listen to books, which makes audio a real competitive necessity now, not a bonus, but a brand-new author with a small or nonexistent fan base simply doesn't have the sales volume to justify, or the income to front, a human narrator's rate.
So the actual choice most indie authors are making isn't "AI assistance versus a human professional." It's "AI-assisted editing or AI narration versus an unedited manuscript or no audiobook at all." Treating all AI involvement as a single, undifferentiated warning sign, rather than separating assistive use from wholesale generation, penalizes exactly the authors with the fewest alternatives to begin with. A traditionally published author with a six-figure advance and a publisher-funded support team was never choosing between AI and nothing. An indie author very often is.
What I Keep Coming Back To
None of this settles whether any specific accusation is true, including this one. But the standard currently being applied wasn't built with this distinction in mind, and until it is, the authors most likely to get caught in the blast radius of a vague AI panic aren't the ones using AI to replace their own writing. They're the ones using it because the alternative, the human expert every warranty clause silently expects, was never truly accessible.
Sources:
- Reported via The New York Times, relayed through multiple outlets covering the Shy Girl cancellation
- Authors Guild, Model Trade Book Contract (Representations and Warranties; Generative AI Rights and Model Publishing Contract Clauses): https://go.authorsguild.org/contract_sections/14 and https://go.authorsguild.org/contract_sections/36
- Miblart, "How Much Does It Cost to Self-Publish a Book in 2026": https://miblart.com/blog/how-much-does-it-cost-to-self-publish-book/
- Authors Republic, "Your Audiobook Production Budget": https://authorsrepublic.com/learn/blog/64/your-audiobook-production-budget
Discussion about this post
No posts


