Novelists Now Need Receipts to Prove They Wrote Their Own Books

Pangram, an AI-detection tool, flagged 78 percent of the horror novel Shy Girl as machine-generated. In March, Hachette Book Group canceled the book's upcoming US publication run and discontinued it in the UK, where it had already been released. The author, Mia Ballard, denies writing with AI. She blames an editor for AI-assisted changes to a previously self-published version and says she is pursuing legal action.

That, in miniature, is the problem now facing professional writers. The same technology they fought as plaintiffs has turned around and pointed at them.

Who is caught in the middle?

Andrea Bartz knows both sides of this fight. Since 2019 she has published five thrillers, hit the New York Times bestseller list, been optioned by Netflix and Hulu, and earned a Reese Witherspoon book club endorsement. She was also lead plaintiff in the class action against Anthropic, which alleged the company downloaded millions of copyrighted books from pirate sites as training data. Anthropic agreed last year to pay $1.5 billion — the largest copyright settlement in US history.

Bartz watched the Shy Girl scandal unfold with a specific dread. "I kind of immediately felt like this was going to lead to a potential witch hunt," she says. AI in publishing, she argues, will "have the unintended effect of leading anyone to be able to accuse anyone of using AI," and authors currently have no way to clear their names. "There's no agreed upon standard," she says, for proving human authorship.

Since Shy Girl, two more fiction authors have drawn AI suspicions — one after winning a short story prize, another after publishers bid into the millions for his crime novel. Both denied the allegations.

Can an app prove you wrote your book?

A crop of startups thinks so. Workings, Blindproof, Wintertext, Human Ink and ObraVera promise to track and document a writer's process, sometimes encrypting text or typing rhythms with unique codes. The aim is twofold: a paper trail against LLM accusations, and provenance evidence when text gets scraped into training sets without attribution.

Mary Rasenberger, CEO of the Authors Guild, says the guild can't endorse any single tool but wants writers to know they exist as a kind of security net. Umair Kazi, the guild's director of policy and advocacy, notes that many of these companies have pitched the guild directly.

Rasenberger's enthusiasm is, shall we say, contained. It's "insulting" and a "burden" to require authors to preserve receipts, she tells WIRED, and "we don't feel it should be required." Her logic is simple economics: "If what you're selling are your words and your voice and your imagination, your creativity, using AI doesn't make sense." Editors she has spoken to say they never see AI-generated submissions — "I'd know it if I did."

How reliable are the detectors?

Not reliable enough to hand them a gavel. Pangram claims an "industry-leading" false positive rate of 0.01 percent on creative writing. Rasenberger says the Authors Guild has "heard of lots of instances where people have put their own human-authored work in Pangram and gotten a result back saying it was AI-generated." Multiple factors, she adds, can fool a detector. She also urges authors to "refrain from accusations of other authors because it just does not support a healthy literary community."

The process-tracking tools have holes too. Max Spero, Pangram's cofounder and CEO, points out that someone "could hand-type AI text, or use an auto-typer." Real writing is messy: "The average writer doesn't just sit down and type out a book," he says. "They work with multiple files, send drafts back and forth, use voice notes while on the go. Not all of this gets captured in writing process tools."

Proof of ownership, not proof of humanity

Brady Ridgway, technical founder of ObraVera, is candid about the limits. His platform can cryptographically "fingerprint" writing patterns and audio recordings; with a time stamp and crypto hash, "if the content crops up later somewhere it's not supposed to be, or somebody else claims it as theirs, we can search the database and say, 'These words were created at this time, by this person.'"

That proves ownership, not honesty. Whether the registered user employed AI remains a self-declaration. Ridgway cites the May kerfuffle over the writer accused of winning a Commonwealth Short Story Prize with an AI-generated submission: "Everybody seems to agree that he's the person who submitted the short story. That's all that the software would do." It can't confirm the declaration is true.

Will publishers ever agree on a standard?

Jane Friedman, author and publisher of the industry newsletter The Bottom Line, doubts the scandals will even last — detectors are improving, and authors have now watched what happens to those who pass off AI writing. She tells WIRED that agents and publishers won't adopt verification that "costs extra money or extra time," and predicts that "it will be like herding cats to get agreement on this, because publishing is slow to adopt new technologies, especially new technologies developed by industry outsiders." Even a shared standard, she argues, would face logistical hurdles and "faked or manipulated" certification. "If any of this costs extra money or extra time, forget it."

Meanwhile, readers are playing detective badly. Bartz has heard of authors getting "review-bombed on Goodreads because some stranger felt that the book sounded like AI to them, and everyone else piled on." Some readers now vow not to read anything published after 2022. "That's a great way to tank an entire industry all at once," she says.

There is an irony pending. Bartz's 2027 thriller, Delete Your Account, is about generative AI — a blocked writer lets a chatbot imitate her favorite thriller and gets trapped inside it. To write it, Bartz produced passages deliberately designed to sound like LLMs. "I don't know — if any of that comes back as AI-written, what will I do?" she says. The detectors, one suspects, will eventually answer that question for her.