At least three of the Big Five US publishers, including HarperCollins, Simon & Schuster, and Hachette, have quietly incorporated generative artificial intelligence (AI) tools into their publishing workflows, even though the practice has not been publicly disclosed and authors' consent was not obtained. These uses involve emailing literary agents, writing promotional and back-cover copy, designing cover art, and generating marketing materials — often at management's direction.
US Publishers Lack Transparency in AI Tool Usage
According to multiple editors, the widely reported cases of authors allegedly using large language models (LLMs) to write books are merely the tip of the iceberg. One editor said every publishing imprint has a story about "some fairly prominent person being pulled for flagrant AI use," but none have been reported in the media. At HarperCollins, for example, a current employee revealed that senior leadership purchased licenses for Anthropic's Claude months ago, but "quickly realized they didn't know how to implement it." Dozens of staff at all levels were then "voluntold" to serve as "AI champions," tasked with surfacing use cases at monthly company-wide brainstorming sessions.
This spring, when some employees raised concerns about AI's legal, ethical, and environmental impacts at these meetings, facilitators dismissed their reservations as "the cost of doing business." The employee portal subsequently added a statement noting that while AI has environmental impact, it accounts for only a tiny fraction of most people's digital carbon footprints. One employee bluntly stated that the company had heard the business case for paying staff decent wages for years, yet poured money into software most people don't want — in 2023, entry-level employees at the Big Five New York publishing groups and Scholastic earned an average annual salary of US$47,583 (approximately HK$370,000).
In addition to Claude, HarperCollins also purchased licenses for ChatGPT and the marketing platform Jasper.
AI Applications in Publishing Spark Widespread Controversy
Feedback from various quarters indicates the most common current use case is generating promotional copy — the text readers encounter everywhere on Amazon, Goodreads, and back covers, used to pitch books to booksellers and media. A publicist at one of the Big Five publishers considers this short-sighted, because media outlets have very low tolerance for AI-generated pitches. Hub City Press publisher Meg Reid noted that big publishing houses using LLMs to write marketing copy is "almost inevitable," and this will become a dividing line between efficiency-driven corporate publishers and smaller presses that insist on "people as the focus."
Additionally, earlier this year, literary agents suspected certain editors were using AI to write rejection letters, because multiple emails from the same editor bore telltale AI writing hallmarks such as "not this, but that" phrasing and three-paragraph structures. Staff at three publishers also believe these suspicions are well-founded. Agents are further worried about manuscript security and have added AI clauses to contracts prohibiting the submission of manuscript content to large language models; HarperCollins' legal team also once barred editors from pasting full manuscripts into open models.
A Hachette spokesperson responded that the company supports operational AI uses that help books reach more readers, but does not support creative uses, including communication with authors and partners. Simon & Schuster global marketing communications officer Wibke Gruetjen stated that employees may use a limited number of vetted enterprise-grade AI tools, but use is not mandatory. Staff-level pushback is also escalating. After Simon & Schuster employees learned the company was piloting monitoring software Skan AI at the urging of its acquirer KKR, they launched a co-signed open letter to CEO Greg Greeley demanding a commitment never to introduce such tools.
Greeley subsequently confirmed in an internal memo that the program was only in early exploratory stages and no adoption decision had been made, but AI detection platform Pangram indicated that at least portions of the memo were written by AI.
Multiple industry figures, including agent Emma Dryden, formerly of HarperCollins and Penguin Random House, said they at minimum want publishers to provide clear guidelines and disclosure around AI use. Her stance is: even if it can't be banned, everyone should communicate more.