For a writer, admitting to using artificial intelligence can now sound less like disclosing a tool and more like confessing to an offence. Say that AI helped you brainstorm, test a plot or tighten a troublesome paragraph, and an uncomfortable question can follow: how much of the book is actually yours?
It is a question publishing needs to confront, but the language surrounding it has become surprisingly crude. A writer who asks an AI tool to suggest research questions is not doing the same thing as someone who generates a chapter. A novelist who uses it to identify continuity errors is doing something different again. Yet all can find themselves placed under the increasingly loaded description of an author who “used AI”.
The danger is obvious. If disclosure carries an automatic suspicion of dishonesty, writers have an incentive not to disclose anything at all.
What Does ‘Using AI’ Actually Mean?
Publishing needs a better vocabulary. An author might use AI to organise research, interrogate the logic of a plot, check whether a character’s motivation makes sense or identify repetition across a manuscript. Another writer might ask it to rewrite paragraphs. Someone else could generate thousands of words and present the resulting prose as their own. Treating every one of those acts as equivalent makes meaningful discussion almost impossible.
Publishing policies themselves demonstrate how difficult the distinction has become. NYU Press permits specified forms of AI-assisted research, brainstorming and editing while prohibiting AI-generated manuscript text. Oxford Academic requires disclosure when generative AI has been used in preparing books. Elsevier distinguishes basic spelling and grammar assistance from more substantive generative intervention. The industry is not dealing with a single question about AI. It is dealing with a spectrum of involvement.
When Suspicion Becomes Part Of Authorship
The consequences of that ambiguity are becoming more serious. The recent controversy surrounding Haitian-Canadian novelist Thélyson Orélien offers an unusually stark example. His novel ‘C’était ça ou mourir’, which had won the Prix du Roman Fnac, became the subject of allegations involving AI-generated writing and plagiarism. Orélien denied using AI to write the book. The Académie Goncourt subsequently removed the novel from consideration, citing concerns that included AI analyses alongside plagiarism allegations.
The case also exposed a greater difficulty: AI detectors do not necessarily agree. Different detection systems reportedly produced different assessments of Orélien’s prose. The author, meanwhile, argued that characteristics identified as suspicious, including repetition and particular rhythms, reflected Haitian and Caribbean linguistic traditions.
Whatever conclusions are ultimately drawn about one novel, the wider problem remains. Once software can contribute to suspicion about whether prose is human, writers face an entirely new form of scrutiny.
Transparency Cannot Work Through Fear
Readers have a legitimate interest in knowing whether a book sold under an author’s name was substantially generated by a machine. Publishers have equally legitimate concerns about copyright, originality and accountability. But transparency requires something more nuanced than a box asking whether AI was used.
Imagine two novelists. One spends three years writing a manuscript and occasionally asks an AI tool to challenge plot holes. The other generates most of a novel through prompts and edits the output afterwards. Technically, both have used AI. Creatively, the resemblance ends there.
Publishing therefore needs to ask better questions: What did the technology contribute? Where did it enter the process? Did it assist the writer’s thinking, alter the writer’s prose or produce the prose itself? Because if every admission of AI use invites doubts about authorship, writers will quickly learn the safest answer. They will simply stop admitting it.