T he romance novelist Coral Hart got her break writing books for Mills & Boon, the publisher of bodice-ripper paperbacks. In 2012, she decided to dip her toe in self-publishing. She felt drawn to dark romance, a subgenre featuring twisted, toxic love stories that was taking off online but that publishers didn’t want to touch. Three years later, things were going well enough that she decided to self-publish full-time.
Kindle Unlimited, where Hart sells her books, rewards prolific writers: the algorithm favours those who can produce a new book at least once a month. Hart says she knows of authors who hire more than a dozen ghostwriters to keep pace. By herself, Hart can write 10,000 words a day, but AI could help her produce many more. Since 2025 she has written 370 novels under 21 pen names using AI, earning herself “high six figures”.
Like any Kindle Unlimited author, she is only paid per pages read: it would make no business sense for her to churn out books that readers quickly abandon. “A bad writer or a bad storyteller is not going to get a good book out of AI. You still need to understand craft. You still need to understand structure. You still need to know what a reader wants,” she tells me, speaking via videocall from her home in South Africa. “So, it’s not taking the human out of the equation. It’s taking the friction and the time and the business blocks out of the way.”
One study (awaiting peer review), led by Tuhin Chakrabarty of Stony Brook University, estimates that around one fifth of recently self-published books contain at least 25% AI-generated text. Hart now teaches courses on how to use AI to produce books. Unless disclosing the use of AI becomes a legal requirement, she advises other authors not to do so because of the public response.
“The minute someone discloses [AI use] they get death threats, they get called names, they’re ousted,” she says. Hart is one of her pen names. After she talked to the New York Times earlier this year she was doxed and faced death threats. “I think the anger and hate comes from a place of fear,” she says. In the literary world, many people fear the tech will undercut and replace them. “I think people are very afraid of AI, and people in publishing are delusional in thinking that it’s not going to overtake our industry, that their screaming on social media is going to stop it.”
But the anger comes from another place, too. If large language models can spit out books that people want to read, or produce a convincing pastiche of your favourite writer, that’s almost certainly because they were trained on established writers’ stolen work. Last year, a class action lawsuit involving around 500,000 titles won a $1.5bn settlement from the AI firm Anthropic for alleged use of the writers’ work without consent. Other similar lawsuits are ongoing. Many writers oppose AI for both moral and aesthetic reasons. One group of 70 prominent authors wrote an open letter calling on the big five publishers – Penguin Random House, Hachette, HarperCollins, Simon & Schuster and Macmillan – to promise that they will “never release books that were created by a machine”. The publishers did not promise.
So where do we go from here? Will big name writers win the argument for keeping AI out of literature, or will our bookshelves and e-readers soon be crammed full of AI-generated stories? And if a reader can’t distinguish human from machine writing, does it still matter?
One literary agent told me that when it comes to AI, the atmosphere in mainstream publishing is like “the wild west”. The technology is outpacing the industry’s capacity to reckon with what it means that authors can – and already are – using AI to write their books. “We’re at a time where it’s all very much in flux, and I don’t think anyone has good answers – but certainly everyone’s talking about it,” says Liz Gately, a New York-based book scout.
Instead, there has been a spate of cancelled high-profile book deals and accusations against writers. In March, Hachette withdrew the romance novel Shy Girl from publication over evidence the author Mia Ballard used AI. Ballard claimed to the New York Times that an editor she hired may have been responsible for inserting AI-generated text. This year’s winner of the Commonwealth short story prize, Jamir Nazir, was accused of using AI , after readers – struck by odd phrases such as “she had the kind of walking that turned benches into men” – ran the story through Pangram, a leading AI detection program. He denies it. Pangram subsequently tested other entrants and concluded that since 2025, four regional winners were likely to have used AI.
In July, agents representing the debut novelist Jerry Falade withdrew from an eye-popping $2m (£1.5m) book deal after he was accused of using AI. He denied using AI for anything other than research. And then, last month, the Canadian-Haitian author Thélyson Orélien’s novel was struck off the Goncourt Academy longlist, France’s most prestigious literary prize, after social media users ran his novel through AI detection software. He, too, denied using AI to write his novel, but did admit to plagiarism in some of his other writing, including a short story he wrote more than a decade ago. After conducting its own investigation, his publishers Boréal dropped him in late October.
Falade – who like everyone I contacted in relation to these scandals declined to speak to me on record – wrote an essay pointing out that writing by non-native speakers or in “culturally unfamiliar” English is more often treated with suspicion. He also suggests the accusations against him may be tinged with racism. “I haven’t seen this kind of investigative zeal directed at white authors accused of using AI …” he writes, “apparently that level of scrutiny is reserved for a Black boy from Nigeria.”
It is true that every writer at the centre of these AI scandals has been an author of colour. In May, the author Steven Rosenbaum – who is white – was found to have accidentally inserted fabricated quotes into his book The Future of Truth: How AI Reshapes Reality when he used AI, which he disclosed in the acknowledgements. He said he regretted the errors but added that it “serves as a warning about the risks of AI-assisted research and verification”, which is why he wrote the book. His book wasn’t withdrawn.
One big problem is that when writers are accused, it is hard for them to clear their names. There is no definitive proof either way. Literary agents and book scouts are professional readers and say most AI-generated submissions (and these days they receive many) are very easy to spot. But, clearly, human judgment isn’t flawless, and evidence suggests that non-native and neurodivergent writers are more likely to raise false suspicion. Academic research estimates that Pangram is at least 95% accurate at detecting AI-generated writing, and it has near-0% rate of false positives – suggesting it is extremely unlikely that the tech would falsely label human-generated writing as AI.
Hannah Bowman, chair of the Association of American Literary Agents’ (AALA) AI special committee, says there’s disagreement on the way forward: some agents think AI detection software should be used routinely, others advocate for relying on relationships of trust. “I think it’s a shame that all of these things are getting addressed so publicly,” she says, with many scandals starting when journalists or social media users run work through AI detectors. “I worry for authors when they don’t have a chance to make a case in defence of their work … there needs to be a fair and thoughtful process, where if publishers and agents have concerns, they can discuss them with the authors.”
In the meantime, Gately, the book scout, says many agents have started recommending their writers don’t send documents as PDFs and save multiple drafts of their work (rather than resaving the same document) in case they are subject to AI accusations. The pervasive anxiety among agents and writers manifests in strange ways. One agent told me they were shocked to receive a manuscript from a writer they had worked with for many years – and who professed to hate AI – that used suspiciously “bizarre” words. She contacted the writer, who was “horrified”: the problem wasn’t that she’d used AI; it was that she was terrified of being accused of doing so and her efforts to make her prose sound “non-AI” had backfired badly.
Another unresolved question is if and when publishers should withdraw books. If Falade’s book was good enough to impress multiple publishers and attract an astonishingly high, multimillion-dollar advance, does it matter how it was written? (Falade, incidentally, has recently signed with a new agent.)
Thad McIllroy, a publishing expert and author of The AI Book Revolution, says he was surprised when Shy Girl was cancelled. The cancellation was partly his fault. After being alerted to online rumours that the book was written using AI, he decided to run it through Pangram, and then approached the New York Times. “I felt so stupid when [Hachette] cancelled it because it never occurred to me that that would have been the response,” he tells me. He anticipated that the publishers would either clean up the text or issue a disclosure on how AI was used. Shy Girl started out as a highly successful self-published novel, and the AI rumours started before Hachette acquired the manuscript. Once bought, the publishers edited the text considerably. McIllroy says he therefore finds it hard to believe that no one raised AI-concerns internally. Plus, he says, it’s a “really powerful book” that clearly resonated with readers, so why drop it the moment the story hit the media?
Both AI enthusiasts and haters say there’s an urgent need for more guidance from publishers, who hold the purse strings and are therefore the kingmakers in the books’ world. McIllroy says publishers need to be “brave” and publish extensive AI policies, outlining how they use the technology in-house and what authors are and aren’t allowed to do. He wonders if publishing houses aren’t doing this because they are in denial about AI, or if they are worried about alienating their highest-earning writers, who are almost all vehemently opposed to the tech. Tess Callero, a literary agent with Europa Content who describes the prospect of AI-generated fiction as “scary”, says she thinks publishers are biding their time because they want to keep the door open for AI. “If the big five publishers were coming down and saying we will not accept [AI-generated writing] it would be taken more seriously,” she says. “But they’re not, and you have to wonder the reason they aren’t. Is it because they see a future in which potentially these books could be monetised?”
One issue is US copyright law, which specifies that a book must be “human authored” and considers AI-generated text public domain. “Which is why it’s so problematic for publishers – because if it’s AI, they have no protection, they can’t have exclusive rights,” says Natalie Lakosil, a literary agent and chair of the copyright committee at the AALA. US copyright law protects AI- assisted work, but if the text has been produced by a machine it risks legal challenge (even if the copyright office, which asks for little detail, accepts the registration). It seems likely, Lakosil says, that this principle will be challenged in the courts eventually, but she is doubtful that the “human authored” requirement will be substantially changed. “The purpose of copyright protection is to protect human creation, to protect artists, and AI is not an artist,” she says. That said, she does believe the law needs to find ways to accommodate writers who want to experiment creatively with AI. She says publishing is in the midst of a “freakout period” comparable to the advent of e-books, which didn’t kill the market – as many feared – but did permanently change it.
In 2023, Stephen Marche wrote the crime novella, Death of an Author, using three AI programs – it is, to anyone’s knowledge, the first AI-novel to be reviewed by the New York Times. It could only be released as an audiobook, because of copyright concerns. Marche has been experimenting with AI in his fiction since the early 2020s. “Publishing has both become incredibly moralistic and at the same time it keeps getting rocked by these scandals,” he tells me. “The people like me who are very conscious about its use, and who both understand the technology and its limits and are not using it in a cheap way, in a cheating way, we’re totally excluded. But you know the scammers will find a way.” He says he would love to see an “experimental blossoming of creative forms” using AI, as is already happening in the art world.
The way forward, Marche says, is full transparency: there is a “very clear moral line” in labelling what is and isn’t AI generated, just as we currently distinguish between fiction and nonfiction. Even those who hate the idea of AI writing (myself included) might recognise that on rare occasions it can be powerful, precisely when it is not trying to pass as human. In Vauhini Vara’s moving essay Ghosts, for example, she uses generative AI to express things she cannot bring herself to write about her sister’s death. The piece is a conversation between a human and a machine that explores the expressive limits, and possibilities, of both kinds of language.
Most likely, AI-generated books will enter mainstream publishing in some form, and will introduce more slop into the market. But Marche points out that our culture is already awash with dull, derivative stuff. He recently went to see a Hollywood action movie that was so formulaic that although it was not made with AI it may as well have been. “There’s this garbage out there that’s worthless, some of it is AI generated, some of it’s human generated. You’re trying to make meaningful shit in this context and how do you do that? Some of it’s going to be made by people. Some of it with AI,” he says.
In France, a new prize – the Prix Pascal de Littérature IA – was launched last month for fiction writers who are working with AI. It seems probable that some AI-assisted books will be good enough that discerning readers cannot necessarily tell that a human has not written them. A recent study, also led by Chakrabarty of Stony Brook University and awaiting peer review, found that in blind experiments, when expert readers (students on leading MFA writing programmes) compared passages written by their MFA colleagues and those written by AI models trained on established writers, and then fine-tuned, they tended to prefer the AI generated text . Would they still prefer the texts if they knew they were AI? Maybe some people will be happy to read such books, although even the most beautiful AI-generated text is a Frankenstein monster, borrowed feelings and observations about the human condition stitched together by an algorithm that has no human qualities, no capacity to love or to feel. But probably many readers would choose human writing. Every agent I spoke to felt committed to protecting the human. “We’re in the human creativity business, not in the business of giving robots jobs,” says Doug Wallace, managing director at Andrew Nurnberg Associates International, a network of literary agencies. AI is “not going to be a big problem because we won’t let it become a big problem”.
But then again, as a writer of course I would want to conclude there. Right now, to write without the aid of a machine is to swim against a strong cultural tide. I’ve lost track of how many friends, in professions other than journalism, have looked at me with something close to pity when I tell them I would never use AI to write, not only because of my contractual obligations or the reputational risk but because I still think I can offer something better than a machine. And yet, machine-generated language keeps cropping up in the places you’d expect to offer the greatest resistance to slop: not just in book submissions but in leading newspapers and literary magazines. Among the many minor AI news stories to bubble up while I worked on this article? A piece in the Times Literary Supplement on AI and the crisis of the author, which was removed from the website because of – you guessed it – allegations it had been written with AI .
