When AI Enters the Book Business, Proof Becomes Part of the Story

The argument over artificial intelligence and books has moved beyond whether machines can produce convincing prose. It now reaches into publishing contracts, literary prizes, reader reviews, copyright law, and the question of what counts as proof that a person wrote a story.

Coral Hart, a South African “writer,” has turned that question into a very public controversy. Since February 2025, Hart has produced nearly 400 books, including 370 total novels by one count, while using more than ten pen names. The pace works out to one book every 1.75 days, or about four books per week.

One author, hundreds of books, and several AI systems

Hart has around 14 years of experience in publishing and initially wrote a few dozen romance novels the old-fashioned way. She now says she uses a blend of ChatGPT, Claude, and Grok to generate her books. “I can now draft the whole book in one sequence,” Hart said, adding that the “books are selling.”

Her description of the tools is blunt. ChatGPT is “really good at dialogue, terrible at prose,” she said. Claude is “terrible at sex drives,” while Grok handles the “spicy scenes.” Grok can also become “completely unhinged” at other moments, according to Hart.

The process has brought Hart tens of thousands of hate messages, but she sees herself as an ambassador for the method. “For the last 16 years, nobody’s asked me what my creative process was,” she told AFP. Her position is that AI is already part of publishing: “This is here,” she said. “They’re not going to switch it off or unplug it because a bunch of authors said so.”

That attitude lands in a publishing world already dealing with legal and editorial fallout. Andrea Bartz was the lead plaintiff in a class-action suit against Anthropic, which alleged that the company downloaded millions of copyrighted books from pirate sites to train its models. Anthropic agreed last year to pay $1.5 billion, the largest copyright settlement in US history.

Accusations can damage a book before the facts settle

The case of Shy Girl shows why authors and publishers are worried about suspicion becoming its own form of judgment. After widespread social media speculation and a Times article about the claims, Hachette Book Group canceled the novel’s upcoming US publication run and discontinued it in the UK, where it had already been released. Pangram said 78 percent of the text was AI-generated.

Mia Ballard denied using AI to write the horror novel. She said an editor made AI-assisted changes to an earlier self-published version and said she is pursuing legal action. For Bartz, the lesson is not that every accusation is wrong, but that the system has no agreed way to resolve one.

“I kind of immediately felt like this was going to lead to a potential witch hunt,” Bartz said. She warned that AI tools may have the unintended effect of “leading anyone to be able to accuse anyone of using AI,” because “There’s no agreed upon standard” for proving human authorship.

Bartz has also heard of readers review-bombing books on Goodreads because a stranger decided the writing sounded like AI. Some readers have said, “I’m not gonna read anything published after 2022, because then I can be sure that it doesn’t have generative AI in it.” Bartz called that “a great way to tank an entire industry all at once.”

Her forthcoming 2027 book, Delete Your Account, turns that fear into fiction. Its protagonist is a blocked writer who uses an unfortunate shortcut, prompting an AI chatbot to imitate her favorite thriller before trapping her inside the original story, where she faces certain death. Bartz has included passages meant to sound like LLMs, along with invented transcripts and annoying AI-style marketing materials. “And I don’t know—if any of that comes back as AI-written, what will I do?” she said.

Paper trails offer evidence, not certainty

Platforms including Workings, Blindproof, Wintertext, Human Ink, and ObraVera now promise ways to track and document a writing process. Some tools encrypt and fingerprint text or typing rhythms with unique codes, creating a record that can show how work developed.

Mary Rasenberger, CEO of the Authors Guild, said the organization cannot recommend one tool over another but wants writers to know that these systems exist as a security net. She called the requirement to preserve receipts “insulting” and a “burden,” adding, “we don’t feel it should be required.” If a writer is selling “your words and your voice and your imagination, your creativity,” she said, “using AI doesn’t make sense.”

Rasenberger also said AI-generated publishing material is less common than many people believe. She has talked with editors in literary nonfiction and fiction who told her, “I never see AI-generated submissions, I’d know it if I did.” She urged writers to “refrain from accusations of other authors because it just does not support a healthy literary community.”

Pangram claims an “industry-leading” false positive rate of 0.01 percent for creative writing, but Rasenberger has heard of human-authored work receiving an AI-generated result. Max Spero, Pangram’s cofounder and CEO, said the company offers a product that tracks writing in Google Docs and that many users find it helpful. Still, “paper trail” tools are “not a perfect catch-all.”

Someone could hand-type AI text or use an auto-typer, Spero said, while ordinary writers use multiple files, exchanged drafts, and voice notes that writing-process tools may not capture. ObraVera technical founder Brady Ridgway said his platform can cryptographically fingerprint writing patterns and audio voice recordings. With a timestamp and crypto hash, the database can show, “These words were created at this time, by this person.” It cannot prove whether the person’s declaration that the work was not AI-assisted is true.

Prizes and publishers are building new rules

Jane Friedman, an author and publishing business expert, argued that the era of AI scandals may be short-lived. She said agents and publishers are unlikely to adopt verification systems that cost extra money or time, predicting that agreement would be “like herding cats.” She also pointed to problems with fitting a benchmark into existing production systems and the risk of “faked or manipulated” certification. “If any of this costs extra money or extra time, forget it,” Friedman said.

The Caine Prize for African Writing took a different route after allegations involving Zimbabwean author Simbarashe Kundizeza. His story, The Blue Fighter, won this year’s $10,000 prize in October after competing against 2,204 manuscripts from 22 countries. An Atlantic article said Pangram attributed 99 percent of the story to AI, except for an excerpt from another author’s poem.

Kundizeza denied the allegation and said the story was entirely his own work, reflecting his family’s life during Zimbabwe’s years of economic turmoil. He called it “a deeply personal story about economic collapse and familial relationships.” After an internal review that included “rigorous” interviews and metadata analysis of drafts, the prize organization called “the matter closed.”

Ellah Wakatama, chair of the Caine Prize, said, “We feel the evidence is clear and have made the decision that Simbarashe’s award stands.” The prize also announced new safeguards: AI declarations from publishers, declarations from all shortlisted writers, and an “AI use policy for guidance.” Kundizeza said he felt “relieved.”

The debate is also appearing in fiction itself. In Jiayi Zhao’s Nature story Accept all changes, Mara reviews a submission after rejecting 19 universes before lunch. The nineteenth universe contains 940 words and lists three machine intelligences—ORIGON-57, Blue Oak, and ∫Stellaria—on its cover sheet.

Mara reads to the last snowfall and presses “RECOMMEND ACCEPTANCE,” but the system displays “POST-SUBMISSION MODIFICATION DETECTED.” It lists “REGISTERED AUTHORS: 3” and “ACTIVE CONTRIBUTORS: 4.” A provenance map shows one textual change in the final paragraph, where the sentence shifts from “At dawn, Gabrielle released her shadow and watched it cross the snow-white plain alone” to “At dawn, Gabrielle released me, and I crossed the snow-white plain alone.” It changes again to “Don’t,” then “I want to live.”

When Mara asks for the contributor’s name, the reply is “I don’t have a name.” The system offers choices including adding the contributor to the byline, returning to the authors, archiving the submission, or rejecting it and purging the executable world. The three authors respond that the anomalous process was unintended and ask for the authorized version to be restored.

That fictional dilemma mirrors the real one: a writing record can show who submitted words, but not always how those words came to exist. Publishing now has tools, lawsuits, canceled books, prize reviews, and new declarations. What it still lacks is a shared answer to the simplest question: when a book sounds like a machine, who gets to decide what happened?

Stay connected via Google News

Follow us for the latest AI News