When a publisher suspects that a manuscript was written by a machine, what happens next? Increasingly, the answer is a software scan, a verdict, and in some cases a cancelled book. On 13 October 2026, the Authors Guild, the largest professional organization for writers in the United States, published a set of best practices that aims to put a stop to this. The article, titled “AI Best Practices for Publishers: Using AI Detectors and Dealing With Undisclosed AI Use”, argues that mutual trust between authors and publishers requires transparency on both sides.

The Guild’s starting point is a growing fear among writers of being wrongly accused of using AI, whether by editors, readers or imperfect detection tools. According to the paper, publishers have already cancelled several books for ostensibly containing AI-generated text. Authors are asked to be open about their own use of AI. In return, the Guild expects publishers to be equally open, not least about the detectors they use themselves.

1. A written AI policy. Every publisher should adopt a written policy stating which AI uses it permits and which it prohibits. It should either be part of the publishing agreement or be handed to the author for acknowledgement at signing. The policy should encourage authors to disclose AI use and commit both sides to bringing a manuscript into compliance together. It should also clarify what the publisher expects when authors warrant that a work is original and solely theirs. The Guild’s most pointed demand is that no publisher should cancel a contract over AI use unless the author breached a signed contract or policy that clearly prohibits that particular use and makes this prohibition a material term of the agreement.

2. No AI editing. Publishers should commit not to use AI to edit manuscripts, apart from pure spelling and grammar correction. They should also not insert AI-generated content, artwork included, without the author’s express written permission.

3. Transparency about AI and detectors. Publishers should disclose when and why they use AI detectors, and should say so in their written policies. The same applies to any AI use in marketing or sales, including whether a manuscript may be uploaded to an internal AI system and for what purpose. Referring to its April 2026 statement, the Guild says employees should never upload manuscripts to public AI systems, which it considers copyright infringement. Uploads should be limited to closed systems run by the publisher, and only with the author’s consent. Third-party detectors deserve particular scrutiny, because using one means handing the manuscript to an outside AI system. Publishers should therefore know how the service stores and uses submitted text, confirm that it is neither retained nor used for training, and obtain the author’s permission. The Guild adds that such policies can change at any time.

4. No sole reliance on detectors. Neither publishers nor agents should decide on the basis of a detector alone whether a book contains AI text. Even the best tools are not always accurate, and self-reported error rates should not be assumed to hold across time and different works. The Guild puts it bluntly: “One false accusation can ruin an author’s career.” If a detector suggests a breach of the publisher’s policy, the publisher must consult the author.

5. Educated editors and open conversations. Editors should be trained to recognize AI use and to talk to authors when they suspect it. A detector result can be the basis for that discussion, under the conditions described above, but not for summary termination or public shaming. Authors should get the chance to rewrite flagged passages. If an author refuses or neglects to do so, persistent AI use may justify rejecting the manuscript, depending on the contract. Once a manuscript has been approved and published, however, AI use alone should not end the contract unless the contract says so explicitly.

6. Care with special cases. Detectors tend to flag writing by non-native speakers more often, because they measure how predictable word choice and sentence structure are. Writers in a second language often use more common vocabulary and more regular syntax. The same can hold for genre fiction with strong conventions and for deliberately plain styles. Publishers should treat such flags as weak evidence and handle the conversation with sensitivity.

7. No proof of authorship. Publishers should not demand that authors prove they wrote their own books. The Guild opposes any requirement to use tools that track or log the writing process as evidence. It does, however, support authors who choose to keep such records voluntarily, in case they are accused later.

The paper is a position statement by an authors’ advocacy group, not a binding standard, and it reflects US contract and copyright practice. Its core logic travels well, though. Detector scores are probabilistic, the consequences for an accused author are not, and a publisher that cannot explain its own tooling is poorly placed to question an author’s. For publishing houses, the practical to-do list is short: put an AI policy in writing, document which tools touch manuscripts, and make sure editors know that a flag opens a conversation rather than closing a case.