TL;DR: Whether to disclose AI assistance in your book is a practical question about communicating with readers, separate from whether AI use is dishonest. Some uses need no disclosure, some warrant a brief acknowledgment, and the heaviest call for more considerable transparency.
I get tired of the way this topic gets argued. One camp treats any AI use as a confession, the other treats disclosure as an insult, and both skip over the reader, who is the person owed something here. I think the rule fits in one sentence: tell readers what they’d reasonably want to know about who wrote the words, and don’t make a production out of tools that never touched the words.
Is AI disclosure an ethics question?
I want to separate two questions that often get tangled. The ethics question is whether AI assistance on a book deceives readers in any morally significant way. A piece on the deception question covers that ground, and the answer is that AI under the labor-split discipline doesn’t deceive readers because the substance is still the author’s. The disclosure question is separate. Disclosure is about practical communication with readers, and the practical answer depends on what was done and on what readers reasonably expect to know.
The two questions are related, and they still need separate answers.
Your book can be entirely yours and still owe readers a line about how you made it. You can also over-disclose, and I think that’s a real mistake. A nervous paragraph about software in the front matter makes a reader wonder what you’re hiding, when one plain sentence would have done the job.
What are the three tiers of AI involvement in a book?
The more of the prose the machine wrote, the more readers are owed.
Books that use AI fall into three categories that warrant different disclosure approaches.
Tier one is light assistance, where AI was used for transcription cleanup, basic research summaries, and connective sections, with all voice work and real material done by humans. The second tier is meaningful assistance. AI played a larger role in scaffolding, research synthesis, or organizational work. The author and writer still produced all voice-bearing material and key arguments.
The third tier is major generation, where AI generated significant portions of the prose that appears in the book, even after human editing.
The tiers are about the practical question of what a reader should know about how their copy of your book came to exist.
Tier one doesn’t require disclosure because no reader would expect to be told about transcription tools or research databases. The second tier warrants a brief acknowledgment because the production process used a meaningful new tool that the reader might find informative. The third tier requires heavy disclosure because the prose itself was machine-generated, and the reader’s relationship to the text changes based on knowing that.
I use AI every day as an assistant for interview transcripts, session summaries and first-pass outlines, and I still do the writing. That’s tier one work. Nobody reading the finished book ever reads a transcript, and asking an author to disclose that kind of help makes about as much sense as asking for a list of the pens on the desk.
Tier one: no disclosure required
If your AI use stayed strictly inside the labor split where the machine handled transcripts, research orientation, citation formatting, and connective drafts that a human rewrote into voice, you owe readers no specific disclosure. The reason is that this category of tool use is structurally similar to other production tools that books have always used without disclosure. Word processors. Spell checkers. Research databases. Library catalogs. Transcription services. Editorial software. All of it’s been part of book production for decades, with no expectation that authors should tell readers.
The relevant question is whether the tools touched the substance of the book or only its production. Transcription cleanup touches production. Research summaries that the author verified touch production. Citation formatting touches production. None of those affect what the book says or how it sounds, and that’s what readers care about. If your AI use was at this tier, you can publish without any AI disclosure and be entirely defensible if asked. A piece on whether readers can tell covers the related question of detection. That is also relevant here.
Tier two: brief acknowledgment
If your AI use went further but still respected the voice rule, a brief acknowledgment in your book’s acknowledgments page is suitable. The acknowledgment can be one sentence. “Research and structural support for this book included AI tools, with all voice-bearing material written by [me/the writing team].” That’s sufficient.
The reader who cares about how books get made has the information. The reader who doesn’t care can skip the line, which most do.
The acknowledgment is a practical note about tooling, in the same way an author might acknowledge a research assistant or an editorial team. That the tool is new and culturally charged doesn’t change the structural nature of the disclosure. Authors who feel anxious about including this line should consider the alternative. Hiding tier-two use entirely puts you in a worse position if the use is later discovered or asked about. A brief honest acknowledgment in the front matter is durable and produces no follow-up problems.
The worst offender is the author who hides tier-two work and then acts wounded when someone asks. That author turned a one-line acknowledgment into a credibility problem, and every reader who hears about it starts wondering what else got left out.
Tier three: substantial disclosure
If AI generated meaningful portions of the prose that appears in your book, including chapters or real passages that weren’t fully rewritten by a human, you owe readers a more serious disclosure. The disclosure should appear in the front matter, in language the reader can’t miss, and should describe what was AI-generated and what was human-written. That’s a practical matter: readers buying the book have an expectation about who wrote the prose, and that expectation deserves to be set correctly.
Authors at tier three should consider whether this is the book they want to publish.
The tier-three book is meaningfully different from what readers usually expect when they buy a book with an author’s name on it, and the marketing position for such a book is different too. Some authors will be comfortable with that positioning and the right disclosure makes it work. Other authors will reconsider and pull more of the AI work back into the human-rewrite category, moving the project from tier three to tier two.
Either response is fine with me. What I won’t defend is publishing tier-three work as if it were tier-one work and saying nothing. That author sold readers a book on the understanding that a person wrote it, and I think that’s a lie told at the checkout counter.
What is the industry doing on AI disclosure?
Publishing norms on AI disclosure are still forming and vary by publisher.
Some publishers now require contractual disclosure of any AI use in submitted manuscripts. Others require disclosure only of considerable AI-generated content. A third group has no formal policy. Self-publishing platforms increasingly require authors to attest to whether AI was used and how, with the requirements tightening over time. Bookstores and reviewers are starting to ask the question, and audiobook platforms have explicit policies on AI narration.
The trend is toward more disclosure instead of less, on a five-year horizon. Authors planning books today should plan for disclosure norms that may tighten before you publish the book, especially for traditional publishing routes. Inclusion of suitable disclosure now, calibrated to the actual tier of use, future-proofs the book against later requirements.
The cost of including a brief acknowledgment is zero. The cost of not including it and having someone ask about it later is real.
Publishers and platforms are tightening these rules because some people flooded the market with machine-written books and passed them off as the real thing. Every author pays for that now, including the ones who did nothing wrong. Don’t give anyone a reason to lump your book in with theirs.
What should you put in your acknowledgments about AI?
For tier one, nothing about AI is required. The standard acknowledgments page covers the people and resources that helped you, and AI tools at this tier aren’t in that category in the same way a research database isn’t in that category.
A single sentence at the end of the acknowledgments page is suitable for tier two. “AI tools were used to support research, transcription, and structural drafting; all voice-bearing material and core ideas are mine [or: came from interviews with the author, depending on the engagement].” Adjust the wording for your specific use, but keep it short and accurate.
At tier three, more major disclosure belongs in the front matter. Something like “Portions of this book’s prose were AI-generated based on the author’s source material and direction; the author has reviewed and approved all content. The ideas, experiences, and judgment in this book are the author’s.” The exact wording depends on the actual division of labor, and getting it right matters more than getting it short. The AI-assisted book service page covers how this disclosure is handled in projects through my practice.
Readers don’t need a confession from you. They need to know who wrote the words they paid for, and if a machine wrote a meaningful share of those words, they deserve to hear it before they buy. Any author who can’t stomach writing that one sentence should treat the discomfort as a sign the book needs more of their own work in it.
