No detection tool ever established that Jerry Falade’s debut crime novel was written by a machine. His own agency withdrew it from publishers anyway.

Europa Content told publishers in a July letter that it could no longer stand behind the manuscript for “Call Me, I’ll Hide the Body” because it was unable to authenticate how the book had developed from origin to completion. Minotaur, a Macmillan US imprint, had won a 14-way auction with a two-book bid reported at $2.4m, roughly £1.8m, and HarperCollins had taken UK rights. Falade denies using AI to write the novel and has said the allegations against him are racially motivated.

What ended the deal was an inability to prove authorship, not evidence of AI, and publishing has arrived at that test by default. The Wall Street Journal reported on 17 August 2026 that agents are now vetting manuscripts themselves while the largest publishers decline to write AI clauses into their contracts. Marketing teams buying copy from freelancers, agencies and ghostwriters are walking towards the same gap from a different direction. Disclosure law has largely spared commercial copy. The private warranties inside a content contract have not been written yet.

What actually collapsed the Falade book deal?

Europa Content pulled the manuscript in July after telling the publishing professionals involved that it could no longer substantiate that the novel was entirely human-written, according to Publishers Weekly, which saw the email. The agency’s account of the book’s quality did not change. Its account of the book’s origins did.

Falade told the Wall Street Journal that he ran an open-source model locally to research the novel, including questions about which chemicals dissolve a body, and that he did not use it to write the prose. He told The Bookseller that he lost US, UK, translation and screen opportunities within six hours of the rumours circulating.

He also disputes the evidence used against him. Falade told Publishers Weekly he had given nobody permission to feed his manuscript to a large language model, and told The Bookseller that passages from books published in 2012 and 2017 registered as 98% and 99% AI-generated when he ran them through detection software himself. Those tests are his own and have not been independently replicated. No publisher or agency involved has produced proof that the novel was machine-written.

Why will publishers not rely on AI detection tools?

Simon & Schuster has said in public that conclusions about an author’s work should not be drawn from detection tools with a record of false positives. The publisher was defending “Daggermouth”, a self-published dystopian romance whose rights it acquired and which has held its place on bestseller lists. The Atlantic reported in July that a Stony Brook University team led by Tuhin Chakrabarty had run more than 14,000 Kindle ebooks through the detection tool Pangram, and that Daggermouth returned a score of 60%. The paper has not been peer-reviewed. The author, H.M. Wolfe, denies that generative AI belongs in the writing process. Pangram’s chief executive, Max Spero, says third-party studies have validated the tool’s accuracy. The Marketing Helm reported in July on a peer-reviewed Vrije Universiteit Brussel study that found Pangram caught 97.5% of fully AI-generated papers with no false positives, and on The Atlantic’s caution that the tool errs more often than its users assume.

The same Stony Brook study carries the figure most useful to a marketing team. Books written with substantial AI assistance made up 20% of the ebook catalogue the researchers sampled, but 12% of sales and 10% of the bestsellers in their genre. Machine-written volume is entering that market considerably faster than it is finding buyers.

The study does not establish that readers can identify AI text on sight. It establishes that AI-heavy titles are over-represented in the catalogue and under-represented in the money.

What does the missing standard mean for commissioned content?

The Authors Guild has written the clearest definition of human authorship anyone has put on paper, and it runs on trust. Its Human Authored certification, opened to all US-published authors in March 2026 after a members-only beta in January 2025, costs $10 a title, is free to Guild members, and requires third-party identity verification and a signed attestation. Nothing scans the text. The definition is the useful part. Text generated by AI disqualifies a work, while research, brainstorming, outlining, indices and spelling and grammar tools do not. That line, drawn between generating the words and supporting the work around them, is one a content brief can actually be written to.

Publishing itself has adopted no such rule. An executive at one of the five largest publishers told the Wall Street Journal the company had decided against specific AI contract language because the field is moving too quickly, and agents describe a trade operating without an agreed standard: some refuse writers who use AI at all, some run detection software, some avoid asking. Emma Dries of Triangle House Literary said agents and editors are being asked “to essentially be police.”

For marketing teams the exposure sits in ownership. The US Copyright Office concluded in January 2025 that human authorship is required for protection, that prompts alone do not make a user the author of an output, and that only the human contribution is protectable. In the UK, section 9(3) of the Copyright, Designs and Patents Act 1988 currently grants computer-generated works 50 years of protection, and the government’s March 2026 report on copyright and AI proposed removing it while keeping protection for AI-assisted work that reflects human creation. Neither regime asks for a label. Both ask who wrote it.

That is a different question from the one the advertising industry has been answering. The Marketing Helm reported in August that the EU AI Act’s transparency rules leave most commercial copy outside the duty to label, and that the IAB’s disclosure framework treats AI-drafted text as below its threshold. A marketing team can be fully compliant with both and still be unable to say who wrote the asset it paid for.

The Authors Guild certification verifies the author’s identity, not the manuscript. It costs $10 a title and rests on a signed attestation.