The Real Question Isn’t Whether AI Wrote It

For six hours in late July, Jerry Falade had one of the biggest debut deals in publishing. Fourteen publishers had bid on his crime novel, “Call Me, I’ll Hide the Body,” and Minotaur Books, an imprint of Macmillan US, won the auction at a reported $2 million to $2.4 million. Then rumors began circulating that he had used artificial intelligence to write the manuscript. “Within six hours, I had been dropped by my agency, deals were paused, and press releases were being issued,” Falade told The Guardian.

His agent, Marc Gerald of Europa Content, said he initially found Falade’s explanation “clear and credible” and chose not to run the manuscript through AI-detection software. That changed after a July 29 meeting, when Gerald said, “aspects of [Falade’s] story changed.” The agency then withdrew the book, citing the need for “total faith” in the manuscript’s origins.

A similar incident happened earlier, in March, with “Shy Girl,” a horror novel by Mia Ballard. Questions about the book’s authorship had been circulating for weeks before Hachette acted, with readers on Goodreads and Reddit flagging passages they believed bore hallmarks of AI-generated prose, according to The Guardian. A widely shared Reddit thread about the book drew hundreds of comments, while a January YouTube video scrutinizing the writing amassed more than 2 million views. Max Spero, founder of detection company Pangram, ran his own test and reported in a post on X that 78% of the book was AI-generated.

Hachette announced it would cancel the U.S. release and discontinue the U.K. edition, citing its own “lengthy investigation in recent weeks.” Ballard denied personally using AI, telling the paper that an editor she’d hired for the earlier self-published version was responsible, according to The New York Times.

Detectors Can’t Answer the Real Question

Chasing detector scores is chasing the wrong target. Both collapses trace back to Pangram, the detection tool at the center of both controversies. The company puts its own false-positive rate at about 1 in 10,000, a precise-sounding number that still only measures statistical patterns in a paragraph, not who wrote it.

The tool flagged passages of Pope Leo XIV’s encyclical as AI-written and labeled several 2026 Commonwealth Short Story Prize entries, including one published in Granta, as mostly AI-generated, according to the company’s blog post.

Wall Street Journal editor James Taranto called Pangram a “defamation machine” after it mislabeled three of his paper’s own opinion columns as AI-generated. A detector that misfires on a papal encyclical and a Journal editorial page can flag a stylistic pattern. It cannot say who reported the story, who structured the argument, or who judged a draft finally good enough to publish.

The unreliability isn’t new. In 2023, GPTZero flagged the U.S. Constitution as AI-written, according to Ars Technica. The document ratified in 1788, meaning James Madison would have needed to be a time traveler to write it with AI.

The reason traces back to how detectors like GPTZero score text in the first place: a metric called perplexity, which measures how predictable a passage’s word choices are against a model’s training data. Formal, heavily codified language, the kind found in constitutions, contracts and scripture, scores low on perplexity because models have absorbed so much of it, and low perplexity reads to the detector as machine-made.

GPTZero Founder Edward Tian, told Ars Technica the Constitution is “fed repeatedly into the training data of many large language models,” which is precisely why his own tool misreads it. OpenAI shut down its detector that same year, citing a “low rate of accuracy,” according to TechCrunch. Tools that mistook the nation’s founding document for a chatbot’s output three years ago are not credible enough for publishers or readers to depend on.

Authorship Was Never That Clean

Publishing keeps treating “did AI write this” as a yes-or-no test, when the industry has never actually needed one.

Ghostwriters have written memoirs under other people’s names for a century. Co-authors build novels off someone else’s outline and voice. Editors rewrite entire chapters before a book ever reaches a shelf. Nobody pulled a memoir off shelves because a ghostwriter typed the sentences. The name on the cover has always stood for the idea, the point of view, the judgment behind the work, not literal proof that one person alone produced every word. AI didn’t invent that ambiguity.

In a new study, researchers led by Stony Brook University’s Tuhin Chakrabarty and Columbia Law School’s Jane Ginsburg, along with co-authors Xinyue Liu and Paramveer Dhillon, ran full-text AI detection across 14,419 self-published genre-fiction titles sold on Amazon between 2023 and 2026.

The study found books with no detected AI text still lost ground. Their share of total sales fell from nearly 100% in early 2023 to about 60% by mid-2026, and revenue per book dropped in seven of eight genres over the same period. Meanwhile books flagged for substantial AI content took a rising share of the market’s scarcest real estate, going from near zero to 31% of new Top 25 bestseller slots.

Copyright law has already made the pivot the industry hasn’t. The Supreme Court declined in March to revisit Thaler v. Perlmutter, leaving in place the rule that works created solely by AI cannot be registered for copyright, while human contributions within an AI-assisted work still can be, according to law firm Mayer Brown.

The question for publishing, then, is not simply whether AI was used. It is what the author contributed. As AI makes polished prose easier to produce, the value of authorship shifts toward the ideas, experience, judgment and point of view behind the words. Detectors cannot measure that. Publishers will have to.

The post The Real Question Isn’t Whether AI Wrote It appeared first on PYMNTS.com.