Publishers Have No Plan for AI-Generated Errors in Nonfiction Books
Steven Rosenbaum used ChatGPT as a research tool while writing his 2026 book about artificial intelligence. The AI tool delivered quotes that were sometimes useful, sometimes serviceable, and sometimes fabricated entirely. In May, the New York Times found more than half a dozen misattributed or fake quotes in the published book-and neither Rosenbaum nor his publisher, Simon & Schuster, had caught them.
The scandal exposed a structural weakness in nonfiction publishing: the industry has no enforceable standards for AI use, no widespread fact-checking infrastructure, and no contractual language to address the problem.
Publishers Don't Fact-Check, and They're Not Starting Now
Nonfiction publishers are not contractually obligated to fact-check the books they publish. Fact-checking is expensive-between $7,000 to $10,000 per book, depending on length-and publishers won't pay for it. Authors can hire fact-checkers themselves, but modest advances often make that impossible.
When editors raised concerns about AI-generated errors last year, publishers told them the responsibility fell elsewhere. "The publisher can't take on the responsibility of fact-checking or hiring a freelance fact-checker because that shifts the responsibility onto us," one senior nonfiction editor said.
A fact-checker would almost certainly have caught Rosenbaum's errors. Some fabricated quotes were only a Google search away from verification.
No Industry Standard Exists for What AI Usage Is Acceptable
Publishers have no agreed-upon rules about when AI for writers crosses the line from research tool to text generation. Contracts typically require authors to warrant they are the sole author and that the work is original, but most lack language specific to AI concerns.
Authors who use AI for research or outlining sometimes forget or downplay the extent of that reliance. What started as assistance in structuring a book can become AI-generated text in manuscript proposals-without the author fully acknowledging it.
"A lot of authors are well intentioned in their use of AI and don't want to rely on AI to generate work that they would then present as their own," said Todd Shuster, co-founder of the literary agency Aevitas. "But before you know it it's not only that they've looked to AI for assistance, they've actually generated texts that they're including in their proposals or manuscripts."
Agents and Editors Are Improvising Solutions
Some literary agents are considering language on their websites stating they won't represent work in which AI generated text. But without binding agreements or verification mechanisms, these statements carry little weight.
One top literary agent's approach is direct: "If I'm getting you a six-figure book advance, I don't want you to be putting it into ChatGPT. Go to the fucking library."
AI-detection tools exist and have become more reliable, but they're not foolproof. They sometimes flag human writing as computer-generated. Running every manuscript through detection software would add cost-something the publishing industry resists.
The Copyright Problem Publishers Aren't Discussing
Even if authors disclose their AI use, another problem emerges: copyright infringement. Large language models often reproduce significant passages verbatim from published works. If those works aren't in the public domain, the fair-use defense is weak.
"Unless there's full disclosure of this, it's improper. And even when there's full disclosure, there's the risk that you as an author, if you're including AI-generated content, are actually engaging in copyright infringement," Shuster said.
The Responsibility Falls on Authors and Agents
Editors at major publishing houses acknowledge the problem but say no one in the industry is willing to establish clear boundaries. "I've always felt that there should be a more in-depth conversation on what things AI absolutely shouldn't be allowed to be involved with," one editor said. "Nobody seems willing to have that conversation."
Publishers are unlikely to invest in new infrastructure or fact-checking services. The burden will remain on authors to verify their work and on agents to vet their clients' human expertise before signing them.
Rosenbaum's book went through three rounds of proofreading but no fact-checking. He has since hired two fact-checkers to work on a corrected version. The New York Times found four substantive errors in a book about truth-a irony Rosenbaum acknowledged while arguing the error rate was below average for nonfiction.
Your membership also unlocks:
