Staff at three Big Five publishers say LLMs now write agent emails and back-cover copy

Earlier this year, literary agents started to suspect that particular Big Five editors were using AI to write rejection letters, so they checked ^[18]. "Among colleagues, we compared emails on different books from the same editors, and they all had the same classic signs of AI writing, like 'not this, but that' and the rule of three," one agent told Wired ^[18]. Employees at Simon & Schuster, HarperCollins and Hachette told Wired they believe the agents were right ^[19].

In public, American publishers spent the year suing Google for copyright infringement and canceling books by authors suspected of using LLMs ^[4]. Editorial staff told Wired that the publicized cases, such as Hachette's horror novel Shy Girl and Macmillan's crime thriller Call Me, I'll Hide the Body, are a small share of the total ^[20]. "Every single imprint has a story of someone very esteemed who had a book canceled because of blatant AI use," one editor said ^[21].

Wired's sources were more than two dozen workers, all speaking anonymously ^[6]. They describe the tools entering the three houses without public disclosure or author consent ^[5]. A champions program like HarperCollins's assumes staff will find the valuable uses ^[8]. What staff describe instead is the easiest use, pointed outward. Publicity copy is one of the most common uses so far, according to workers at all three houses ^[13]. Staff also describe LLMs designing cover art and generating marketing materials, often at an executive's direction ^[7]. That copy is easy to template. It is also the text outsiders read: it appears on Amazon, Goodreads and back covers, and it is how books are pitched to booksellers, journalists, content creators and critics ^[17].

A Big Five publicist told Wired that "using AI to generate publicity copy is short-sighted." "Publicists have a duty to engage in good faith with readers and our colleagues in the media, who I'm sure have a low threshold for tolerating AI-generated pitches," the publicist said ^[14].

The objections inside come from people on publishing pay. "It was such a slap in the face," one HarperCollins employee told Wired. "We've been making valid business cases for years to be paid a healthy salary, and instead they spent that money on software that most of us don't really want." ^[11] In 2023, entry-level publishing staff in New York City at the Big Five and Scholastic earned $47,583 on average ^[12]. After staff raised environmental concerns, HarperCollins added a line to the AI section of its employee portal: "while AI does have an environmental impact, it's a small part of most individuals' total digital carbon footprint" ^[10].

Meg Reid, publisher of the independent Hub City Press, said LLM-written marketing and publicity "feels pretty inevitable" for the Big Five. "That's one place we're probably going to see a bigger shift between corporate publishers looking for efficiencies and smaller publishers insisting that the humans involved are the whole point," she told Wired ^[15].

Wired's reporting does not include what the licenses cost or whether AI-written copy sells more books. For a team rolling out the same tools, I'd sort each proposed use on two lines. One is whether the text leaves the building. The other is whether it goes out under a person's name or on a person's work, such as an editor's signature or an author's cover. Internal, unsigned uses, like summarizing a meeting, are where a team can try things without asking anyone. Outbound, signed uses, like a rejection letter or a back cover, are where the recipient and the author have a claim to be told. The two mixed squares need a written rule before anyone starts. The forcing function is to write the one-line disclosure you would send the person on the receiving end. If no one at the company would approve sending it, that use does not belong in an undisclosed rollout.