This is a sidebar to Hachette Pulls Novel Due to AI Use
Again and again on social media, I saw writers worried about the presumably insufficient evidence that Mia Ballard used AI in her novel Shy Girl. It bears repeating: The initial evidence arose from readers analyzing the work. Presumably Hachette felt the evidence was overwhelming, but I can’t explain why they didn’t act until the New York Times article. And the Times, for its part, was working off evidence first presented by publishing consultant Thad McIlroy, who recently wrote about his key role in surfacing the story.
Most authors and organizations, including the Authors Guild, do not trust the accuracy of AI detection programs. Such programs deal in probabilities and percentages and are known to generate false positives. Some speculate Ballard could have a legal case against Hachette and/or the New York Times for damaging her reputation without hard evidence of AI use, as there’s no way to prove definitively that AI has been used in a work. However, she confessed to the role of AI herself.
Some authors have argued that because large language models have been trained on their work, their work will be mistaken for AI writing. But that is a misunderstanding of LLMs like Claude, ChatGPT, and so on. Models are trained on vast data sets, and few if any contemporary authors have produced enough work to influence the outputs in a signature manner—unless LLMs are prompted to write in a particular style, which is a topic for another time.
When you see screenshots from AI detection tools showing the US Constitution or Mary Shelley’s Frankenstein as likely to be 100 percent AI generated, you are seeing (a) a poor AI detection product and (b) the repetition and prevalence of AI model training on public domain work. While there is reason to question the accuracy of AI detection tools, just because LLMs are trained on your work doesn’t mean they will flag your writing as AI. AI “tells,” to the extent they exist, have to do with the peculiarities of AI-generated writing. These tells do change over time and perhaps will go away with further development. (What are the tells? Start here.)
I don’t believe any Big Five or commercial publisher currently makes widespread, institutional use of AI detection software to review manuscripts at any stage of the process. But after Shy Girl, they might consider it to help them assess potential problems early on. AI detection is not foolproof, but there are publishers who rely on it to start conversations with contracted authors about AI use in manuscripts. Emphasis on the word start. For some editors, it helps to have a third-party measure to start those conversations because it’s easier than making what feels like a personal accusation. One agent commented on Substack, “The academic side of publishing has been on top of this from the jump, but on the trade side of things, it’s been rather laissez-faire and trusting that authors remain honest.”
Jane Friedman has spent her entire career working in the publishing industry, with a focus on business reporting and author education. Established in 2015, her newsletter The Bottom Line provides nuanced market intelligence to thousands of authors and industry professionals; in 2023, she was named Publishing Commentator of the Year by Digital Book World.
Jane’s expertise regularly features in major media outlets such as The New York Times, The Atlantic, NPR, The Today Show, Wired, The Guardian, Fox News, and BBC. Her book, The Business of Being a Writer, Second Edition (The University of Chicago Press), is used as a classroom text by many writing and publishing degree programs. She reaches thousands through speaking engagements and workshops at diverse venues worldwide, including NYU’s Advanced Publishing Institute, Frankfurt Book Fair, and numerous MFA programs.
