Books by People bets on drafts to certify human-written books

Evidence, not a detector score

Earlier labels such as Human Authored relied on authors to confirm that they had not used AI. Books by People wants to add a review of the evidence behind that claim.

Co-founder Esme Dennis told WIRED that the review assesses how convincingly those documents show a book was written by a person through the usual process of drafting and revision. The UK Intellectual Property Office approved both the mark and the review procedure.

Publishers will pay a fixed membership fee, plus a separate fee for each book they submit. Dennis says Books by People does not describe its process as AI detection: the company concluded early on that it could not keep pace with continually improving AI models. When evidence leaves the company uncertain about AI use, it says it decides in the author’s favour. Its guarantee, Dennis says, is that a book has passed the check—not that the process can establish its origin with certainty.

The trust problem

A series of publishing controversies this year has put the industry’s implicit assumption under pressure: that a book on the shelf was written by a real person who did not plagiarise. A literary agent withdrew a deal reportedly worth $2 million; a publisher removed a novel from shelves; and readers alleged that an award-winning story was AI-generated.

Some people in publishing say they are less worried that AI will replace writers than that readers will not want machine-written books. Science-fiction writer Ted Chiang, whose 1998 novella “Story of Your Life” was adapted into the Oscar-winning film Arrival, has compared reading machine-generated text to watching finished objects fall off a conveyor belt. Knowing how a work was made, he says, is inseparable from the experience of reading it.

Sam Jordison, co-director of independent publisher Galley Beggar Press, which plans to use the mark, says publishing has long relied on trust among authors, publishers and readers. Reading AI-generated text presented as human, he said, feels repellent and deceptive. Former Penguin Random House CEO Markus Dohle called human storytelling “the lifeblood of our industry” in a statement supporting Books by People.

But a mark intended to protect trust creates a different risk: an author wrongly suspected of using AI could suffer lasting reputational damage. Tom West, chief executive of Publishers’ Licensing Services, has warned against a “witch hunt” and says the details of the checking process will matter.

The risk is not hypothetical. Pangram, a New York startup that says its AI-text detection system is 99.7% accurate, estimated that Jerry Falade’s book was 97% AI-generated. In September, Falade, a debut author whose agent reportedly withdrew a $2 million book deal, said the software had treated his Nigerian dialect as evidence of AI use. In a recent essay, he wrote that writing a book can take months or years, while a detector can take two minutes to destroy a life’s work. Pangram did not respond to a request for comment.

What the mark can—and cannot—prove

Some critics argue the burden should run the other way: platforms and publishers should disclose when material is AI-generated, rather than requiring authors to prove they wrote it themselves. West says authors are repeatedly asked to account for where their ideas and words came from, and that tendency should be resisted. A book, he says, should be presumed human-written.

I think that tension is the point of Books by People’s approach. Its process avoids treating a detector score as a verdict, but still asks authors to produce evidence to secure a mark. If the label spreads, the question will not just be whether publishers value human writing; it will be whether they can make that value visible without making authors prove their innocence.

Chiang has compared rules against cheating at universities to a publishing standard against misrepresentation: a policy matters even when no enforcement system can be perfect. The mark may work less as a guarantee about every page than as a statement of what publishers want readers to believe—and what they are prepared to check.