The use of generative AI for promotional copy, work emails and other tasks is reportedly spreading at U.S. publishers HarperCollins, Simon & Schuster, and Hachette. Employees and literary agents are calling for clear policies and for authors to be informed. At Simon & Schuster, employees are also opposing a proposal to adopt a business analytics tool.
Strict on Authors’ AI Use, but Embracing It In-House
The report is based on anonymous interviews with more than 20 people working in the publishing industry. Employees said that at least three of the five largest U.S. publishers use tools such as Claude and ChatGPT for promotional copy, book jacket descriptions, marketing materials, and emails to literary agents.
Meanwhile, the report also describes cases in the publishing industry in which books by authors suspected of using generative AI have been dropped. Publishers take a hard line on authors’ manuscripts while using AI to create writing that reaches readers. How they explain this distinction to authors and readers is at the heart of the issue.
Still, generating the text of a novel and writing promotional copy are not the same thing. Precisely because the uses differ, publishers must spell out what they allow and what they do not; otherwise, people outside the company cannot assess the distinction.
Is Promotional Copy Just an Administrative Task?
One prominent use cited by employees is writing promotional copy for books. Such text reaches readers on online retailers’ sites and on book jackets, and is also used to introduce books to booksellers, journalists, reviewers, and others. Though produced behind the scenes in publishing, its role is to connect works with readers.
Simon & Schuster says employees in each department can use a limited set of enterprise AI tools, but are not required to do so. The company prohibits the use of unapproved tools and says its aim is to free up time for creative work with authors.
Hachette also says it supports the use of AI for business tasks that help bring authors’ books to more readers. However, it does not support creative uses, including communications with authors and business partners. The extent to which this policy aligns with employees’ accounts of how email is actually being written needs further scrutiny.
How Much of an Unpublished Manuscript Can Be Entered?
Literary agents are concerned that authors’ unpublished manuscripts could be entered into AI systems to produce book descriptions or rejection letters. Literary agents represent authors in negotiations over contracts and rights with publishers. One agent interviewed said they had added a contract clause prohibiting manuscripts from being entered into large language models.
Here, it is important to distinguish where input is sent, whether it is stored, and whether it is used for training. Using cloud-based AI alone does not establish that a manuscript was used for training. The material reviewed for this report also does not reveal the details of each company’s contract terms or data-management settings.
At HarperCollins, an editor reportedly asked whether a full manuscript could be entered into an external AI tool, but the legal department did not approve. The existence of concerns is separate from whether a manuscript was actually sent outside the company. No information was presented to substantiate the latter.
When Efficiency Initiatives Turn Into Employee Backlash
At HarperCollins, employees were reportedly asked to attend regular meetings to explore ways to use Claude after the company introduced licenses for the tool. At Simon & Schuster, employees said an internal contest soliciting ideas for AI applications was held in May 2024, with a $10,000 prize for the winner and $5,000 each for two runners-up.
Dissatisfaction with such initiatives is not directed solely at AI’s capabilities. Some employees say that layoffs have increased their workloads, and that they are now being asked to take on the additional burden of finding ways to use new tools after already calling for adequate staffing and better compensation. However, accounts of staff reductions and increased workloads are based on anonymous testimony and do not constitute statistics on the companies as a whole.
There is a gap between companies’ claim that efficiency gains will free up time for creative work and employees’ sense that they have no choice but to use AI to make up for staff shortages. Evaluating these initiatives requires looking not only at time saved, but also at whether employees’ workloads have actually decreased.
Employee Distrust Comes to the Fore Over Proposed Business Analytics Tool
At Simon & Schuster, consideration of a product from Skan AI prompted employee backlash. Employees gathered signatures for an open letter to CEO Greg Greeley, urging the company not to introduce similar software into the workplace. They are concerned that, in addition to monitoring work, its ability to identify tasks that can be automated could lead to staff being replaced.
An AI tool that drafts promotional copy and software that analyzes employees’ work have different implications. The dispute over the latter shows that publishers’ use of AI extends beyond writing assistance to concerns about the distribution of work and job security. However, it has not been established that the current proposal is intended to reduce staffing.
In a memo to employees, Greeley acknowledged that the company is evaluating Skan AI’s Blueprint, but said it has not decided whether to use the product or other productivity-monitoring and work-analytics tools.
Comments