Book publishers are using AI in secret, and employees are pushing back

The quiet rollout

Employees at HarperCollins, Simon & Schuster and Hachette say AI tools are already being used for marketing copy, rejection letters and other work, often without clear rules or consent from authors. The conflict is not simply about whether publishers should use AI. It is about who gets to decide what goes into the tools, what work they replace and what authors and staff are told.

Editors say reports of authors accused of using AI to write books—including Hachette’s horror novel Shy Girl and Macmillan’s crime thriller Call Me, I’ll Hide the Body—may be only the visible part of the story. One editor said every division has a story about a respected author whose book was canceled over apparent AI use, but never made the news.

At HarperCollins, senior leaders bought licenses for Anthropic’s Claude several months ago, then realized they did not know how to put them to work, according to a current employee. Dozens of junior and senior staff were effectively told to become “AI champions,” meeting monthly with colleagues across departments to find uses for the technology.

At one meeting this spring, employees raised legal, ethical and environmental concerns. A participant said the meeting’s host called those objections “the cost of doing business.” HarperCollins later added a page about AI to its internal portal, saying that although AI affects the environment, its impact is small compared with most people’s overall digital carbon footprint.

HarperCollins also acquired licenses for ChatGPT and Jasper, a platform with AI agents that promises to run marketing processes end to end. At least one division has actively used AI since getting access, including to make marketing videos, an employee said. Workers at HarperCollins, Simon & Schuster and Hachette say one of the most common uses of large language models is writing promotional copy.

That copy reaches readers on Amazon and Goodreads, on book jackets and in pitches to booksellers, journalists, content creators and critics. One publicity specialist at a Big Five publisher called AI-written copy shortsighted: publishing staff have a duty to deal honestly with readers and the media, who, in the specialist’s view, have little patience for AI-generated pitches.

The work authors cannot see

Earlier this year, literary agents began suspecting that some Big Five editors were using AI to write rejection letters. Agents compared letters from the same editors about different books and noticed patterns associated with AI writing, including “not this, but that” constructions and the rule of three. Employees at Simon & Schuster, HarperCollins and Hachette told WIRED they considered the agents’ suspicions credible.

A few months ago, Simon & Schuster employees attended a workshop with OpenAI representatives. One attendee said the presenter did not understand what publishing employees do, yet suggested that ChatGPT could be used to write letters to literary agents. Workers say editors at HarperCollins and Hachette increasingly do that.

The concern goes beyond tone. All the literary agents interviewed worried that publishers might upload manuscript excerpts to large language models to draft rejections or promotional descriptions. One agent, who asked to remain anonymous because of career risks, said contracts now include clauses barring publishers from sending authors’ material to large language models. The clauses were initially meant to protect authors from technology companies, but also apply to the publishers that acquire and release their books.

An open model can send an unpublished manuscript to a language model. A closed model, unlike public services such as Google Gemini and ChatGPT, runs on a separate computer or network without connecting to the developer’s cloud servers. A HarperCollins source said several editors had asked managers whether they could upload complete manuscripts to an open model; the legal department prohibited it.

Simon & Schuster also ran an internal contest for AI ideas in May 2024, offering $10,000 to the winner and $5,000 to each of two finalists. The company later promoted AI less after employees complained at all-hands meetings, one worker said. Some employees also say AI is used to create book covers. Some foreign publishers distributed by Simon & Schuster openly disclose AI use when publishing sketches.

Meg Reid, publisher of independent Hub City Press, sees AI-written marketing and publicity copy as nearly inevitable at the Big Five. She expects a widening divide between large publishers seeking efficiency and smaller presses that value the people doing the work.

Efficiency for whom?

Some publishing employees say they are reluctant to try large language models because management has cut their teams and tripled their workload. One described large departments shrinking to five or six people while each person took on twice or three times as many books. The employee did not blame colleagues for looking for shortcuts, but said the responsibility lay with companies pushing those tools on workers under pressure to keep up with competitors.

Simon & Schuster staff also began collecting signatures for an open letter to CEO Greg Greeley after learning the company was experimenting with Skan AI, apparently at the urging of KKR, the investment firm that bought the publisher in 2023. Skan AI says its product helps managers cut costs, decide what to automate and meet board demands for company transformation.

Employees objected to both potential workplace surveillance and the possibility that tools for identifying automatable tasks could be used to monitor and replace them. The letter demanded a promise that Skan AI or similar programs would not be used in the company’s workflows, now or in the future.

On October 6, the day the letter was mentioned in New York magazine’s Book Gossip newsletter, Greeley told employees that Simon & Schuster had evaluated Skan AI’s Blueprint product but had not decided whether to use it or other tools for productivity monitoring and workflow analysis. He said he had not seen materials about the preliminary evaluation before employees raised their concerns the previous day. Pangram, a service used to detect AI-generated text, indicated that at least part of Greeley’s memo may have been written with AI.

My read is that the dispute is becoming less about whether a publisher has an AI license and more about what management can do with it without telling the people whose work and writing are involved. The quietest part of these rollouts is also the most consequential: employees and authors still do not know what information is being put into these systems, or where the boundary lies between assistance and replacement.

The rules are still missing

Hachette says it supports AI for work tasks when it helps authors’ books find more readers, but does not support creative uses, including correspondence with authors and partners. Simon & Schuster’s global marketing and communications director, Wiebke Grutjen, said a limited number of corporate AI tools are available to staff across departments, but their use is optional. The company prohibits tools that have not been reviewed and approved, she said, with the aim of giving employees more time for focused, creative work with and for authors. HarperCollins did not respond to a request for comment.

The policy questions reach beyond staff workflow. Nearly two years ago, HarperCollins CEO Brian Murray said at the UBS Global Media and Communications Conference that AI offered book publishers more opportunities than risks. He suggested it could help HarperCollins publish more translated books and audiobooks. Since then, editors have seen manuscripts translated with large language models and labeled as AI-assisted at the Frankfurt and London book fairs, where US publishers buy publication and translation rights from foreign literary agencies.

In August, Michael Reynolds, executive publisher of independent New York publisher Europa Editions, spoke to booksellers at the New Voices New Rooms conference in Baltimore about his enthusiasm for AI translation. Listeners said he seemed puzzled by criticism of the technology. Europa Editions already draws criticism in literary circles because its book covers do not name human translators—at least, not yet.

Dan Sinykin, author of Big Fiction: How Conglomeration Changed the Publishing Industry and American Literature, said executives at the Big Five resemble leaders of other large US companies more than the editors, publicists and translators who work for them. In his view, management is always looking for ways to increase profit and is now testing what it can get away with before employees revolt.

Most of the professionals interviewed want publishers to set clear AI rules and disclose where and when they use the tools. Emma Dries of Triangle House Literary, formerly an editor at HarperCollins and Penguin Random House, said publishers should clearly explain which tools they use and in what circumstances. A current HarperCollins employee said management had not told staff what to say to authors and agents, perhaps because doing so would provoke outrage.

Dries added that if a complete ban is impossible—though she would prefer one—people should talk more openly about how the tools are used. That is the tension publishers have yet to resolve: a tool sold as a way to free up time can also make it easier to shrink teams, process authors’ work without their knowledge and keep the new rules unwritten.