Book Publishers Are Quietly Using More AI. Staff Are Revolting

A large language model, or LLM, is a computer system trained to recognize patterns in written language. It generates text by predicting which words or pieces of words are most likely to follow a prompt. The result can resemble writing by a person, even though the system does not have human intentions or experiences.

Axiom and Paradox are examples of tools built around LLMs. A user supplies an instruction, such as drafting publicity copy or an email. The model analyzes that instruction and produces a response by selecting likely language sequences. Users can then revise, reject, or reuse the output.

The article reports that HarperCollins bought Paradox, Axiom, and Jasper licenses. Workers used these tools for publicity copy, marketing materials, emails, cover art, and marketing videos. The article does not explain the models’ technical training, but it shows how publishers are applying their text-generation abilities.

At least three Big Five US publishers named in the article are using AI: HarperCollins, Simon & Schuster, and Hachette. Their companies are applying tools such as Paradox and Axiom to routine publishing and marketing work. The activity has often happened quietly, without public disclosure or author consent.

Workers describe LLMs generating emails to literary agents, publicity copy, back-cover copy, and other marketing materials. HarperCollins also bought Jasper, which promises end-to-end marketing workflows. At least one HarperCollins division has used AI-generated marketing videos. Agents also suspected some editors used AI to write rejection letters.

The most common reported use is publicity copy. That material appears on Amazon, Goodreads, book covers, and sales pitches to booksellers, journalists, critics, and creators. Hachette says it supports operational AI uses but not creative uses or AI communication with authors and partners. Simon & Schuster says approved enterprise tools are available, but optional.

The article documents AI adoption at HarperCollins, Simon & Schuster, and Hachette, three of the Big Five US publishers. It does not provide a company-wide percentage or total employee count. Instead, the evidence comes from interviews with more than two dozen workers and from staff reports about licenses, projects, and internal programs.

At HarperCollins, senior leaders bought Paradox licenses and then asked dozens of junior- and senior-level employees to become “AI Champions.” They held monthly brainstorming sessions to identify practical uses. Staffers also said the company purchased Axiom and Jasper licenses. One division was described as “all in,” including the production of AI-generated marketing videos.

Adoption therefore appears broad but uneven. Simon & Schuster gives employees access to a limited number of approved enterprise tools, without making them mandatory. Hachette supports some operational uses but rejects creative uses. The article suggests expansion is underway, even as disclosure and consent remain inconsistent.

Employees are concerned about AI’s legal, ethical, and environmental effects, as well as its impact on pay and creative work. Some HarperCollins staff felt insulted that the company spent money on software while employees had spent years seeking healthier salaries. A meeting leader reportedly dismissed objections as “the cost of doing business.”

Literary agents are worried about trust and authenticity. They compared rejection emails from the same editors and noticed repeated patterns associated with AI writing, including “not this, but that” and the rule of three. Publicists also warned that AI-generated pitches could damage good-faith relationships with readers and the media.

Authors face a consent problem because publishers may use AI in processes connected to their books without informing them. They also face inconsistent standards: publishers publicly oppose some creative AI uses while quietly encouraging operational experimentation. The article reports that several books were canceled after suspected or blatant AI use, sometimes without public disclosure.

The most serious reported consequence is cancellation. Multiple editorial staffers say publishers have canceled books after discovering blatant AI use, even when the cases did not reach the news. One editor said every imprint had a story involving an esteemed person whose book was canceled for that reason.

The article names two public examples. Hachette canceled the horror novel Shy Girl, and Macmillan canceled the crime thriller Call Me, I’ll Hide the Body, after authors were accused of using LLMs to help write them. The source does not describe every contractual or financial consequence for those authors.

The response is not uniform across publishing. Companies publicly present different boundaries for AI. Hachette says it permits operational uses that help books reach readers but rejects creative uses. Simon & Schuster allows approved enterprise tools without requiring employees to use them. These policies show that discovery can trigger serious action, but standards remain uneven.

Tasks likely to remain primarily human include editorial judgment, creative writing, relationship-building with authors and agents, and meaningful communication with media partners. The article shows why: publicists describe good-faith engagement as part of their duty, while agents value personal and trustworthy rejection messages. Hachette explicitly says it does not support creative AI or AI communication with authors and partners.

Companies are trying to automate more routine production work. Reported uses include drafting publicity and back-cover copy, emailing literary agents, creating marketing materials, generating cover art, and producing marketing videos. HarperCollins also bought Jasper, a platform promising to run end-to-end marketing workflows.

The balance is unsettled. Some executives see AI as a way to create efficiencies and free employees for deeper work. Other workers see it as a threat to salaries, standards, and the human purpose of publishing. Meg Reid says large corporate publishers may shift toward automated marketing, while smaller presses may emphasize human involvement.

Copyright concerns whether protected writing, images, or other material can be copied, adapted, or distributed. Authorship concerns who created the expressive work and who receives credit or control. Consent concerns whether an author, employee, or partner agreed to AI involvement. In practice, contracts, company policies, and applicable law help determine what publishers may do.

The article shows these issues colliding. Publishers are using AI for publicity copy, cover art, emails, and marketing without public disclosure or author consent, according to workers. It also describes authors whose books were canceled after alleged LLM use. Meanwhile, publishers have publicly opposed some AI practices while adopting others internally.

The source does not state a definitive legal test for every use. It reports that HarperCollins employees raised legal concerns, and that publishers are setting different boundaries. Any decision about AI-generated text or images therefore depends on rights, contracts, permission, and the publisher’s own rules, rather than on one universal policy described here.