Should Authors Disclose AI Use in Books? Publisher Policies Compared

You finished the manuscript, you’re staring at the upload page, and there it is: a question asking whether you used AI. Suddenly a simple “yes or no” feels like a trap. Do you have to say anything? And if you do, will it kill the book before anyone reads a word?

The honest answer is that it depends entirely on where you’re publishing, and the major players don’t agree with each other. Let’s compare what they actually require.

Key takeaways

• There’s no single rule. Whether you must disclose AI use depends on where you publish, and the major platforms now disagree on the details.
• Amazon KDP requires authors to declare AI-*generated* content during upload, but treats AI-*assisted* work (where you create it and use AI to refine) differently and doesn’t ask you to disclose that.
• KDP’s catch: under its own definition, AI-generated content stays generated even after substantial edits. Editing the output doesn’t reclassify it. What matters is who created the content in the first place.
• Most traditional publishers don’t have a public disclosure checkbox, but they increasingly bake AI clauses into author contracts you’re expected to honor.
• Disclosure is about transparency, not punishment. Declaring AI-generated content to a platform usually doesn’t block publication.

First, the question hiding inside the question

When people ask “do authors have to disclose AI in books,” they’re usually blending two different things: a legal obligation and a platform rule. They’re not the same.

There’s no general U.S. law that forces a novelist or nonfiction author to slap an “AI was here” label on a book. So in the strict legal sense, no, you’re not breaking a statute by staying quiet. But that’s rarely the obligation that bites you. The rules you’ll actually face come from the company you publish through. Those are private agreements, and you accept them the moment you upload or sign. Break one and a title can get pulled or a contract voided, no courtroom required.

So the useful version of the question isn’t “is it the law?” It’s “what did I agree to?” And that changes depending on the door you walk through.

The dividing line: generated vs. assisted

Before comparing platforms, you need the one distinction that drives almost every policy: AI-*generated* versus AI-*assisted*. Amazon KDP made this split mainstream, and it’s worth learning because other publishers borrow the logic.

• AI-generated means an AI tool created the content (text, images, translations). KDP is explicit that this label applies even if you edited the output substantially afterward. Prompting a model for a chapter and pasting it in lands here, and so does prompting a chapter and then reworking it heavily.
• AI-assisted means *you* created the content and used AI tools to brainstorm, edit, refine, or error-check along the way. You wrote the chapter; a model helped you tighten it.

So the test isn’t how much you edited. It’s who authored the words to begin with. Why does that matter? Because most disclosure rules only target the first category. The platform doesn’t care that you ran your own prose through a grammar tool. It cares whether a machine wrote the book and you’re presenting it as authored work.

That line tends to track something on the page, too. Heavily generated text often carries the statistical fingerprints automated tools look for. If you want the mechanics, we’ve written about the statistics behind how AI detectors read text. The honest point: genuine human authorship is what keeps a manuscript on the “assisted” side of the line, in both the policy sense and the way the words actually read.

Amazon KDP: an explicit declaration, quietly handled

KDP is the clearest case because it spells the rules out. When you publish through Kindle Direct Publishing, the submission flow asks you to declare whether your book contains AI-generated content, broken out by type: text, images, and translations. You answer for each.

Crucially, KDP asks you to declare AI-*generated* material, not AI-*assisted* material. If a model drafted passages and you’re publishing that as your text, that’s a declaration, and KDP says editing those passages doesn’t change the answer. If *you* wrote the book and used AI to polish, KDP’s guidance treats that as assisted and doesn’t require you to flag it.

Two things people get wrong about this:

First, the declaration isn’t published to readers. It’s information for Amazon, not a banner on your product page. So the fear that disclosing will scare off buyers is mostly misplaced; nobody browsing your listing sees a checkbox you ticked during upload.

Second, declaring doesn’t automatically block your book. The KDP declaration is informational. What actually gets titles removed is abuse: flooding the store with high-volume, low-effort uploads, or violating content and intellectual-property rules. In September 2023, around the same time it added the AI disclosure requirement, Amazon also capped most accounts at three new title uploads per day to curb spam. Honest disclosure of generated content is not the thing that triggers a takedown. Hiding generated content you were asked to declare is the riskier move, because it’s a policy violation on top of whatever else.

Mini-scenario: same tool, two authors, two answers

Priya writes a cozy mystery over six months. She uses a chatbot to brainstorm red herrings and an editing tool to catch repetition, but every sentence in the final book is hers, written and rewritten in her voice. At upload, she truthfully answers that the book is not AI-generated. Her use was assistive, and KDP doesn’t ask her to declare it.

Marcus prompts a model to produce a “30 quick recipes” ebook, runs a spell-check, and uploads it three days later. The text is the model’s. Under KDP’s definitions, that’s AI-generated content, and he’s expected to declare it. Notice that even if Marcus had spent a week polishing those recipes, KDP’s rule wouldn’t budge: a model wrote them, so they stay in the generated bucket. Declaring it isn’t the problem. The problem would be saying “no” and getting caught, or stacking a dozen near-identical titles in a week.

Same tools, opposite obligations, decided by who actually authored the words.

Traditional publishers: fewer checkboxes, more contracts

Walk through the traditional door and the landscape looks different. Most trade and academic publishers don’t have a public disclosure form the way a self-publishing platform does. There’s no upload wizard asking you to tick a box.

What’s changed is the paperwork. Publishers have started writing AI clauses directly into author agreements. A typical clause asks you to *represent and warrant* that the work is substantially your own and to disclose, or outright limit, AI-generated material. Some require that the manuscript contain no AI-generated text without prior written consent. So you may be agreeing in writing to disclose, even without a visible button anywhere.

This is why “my publisher never asked” is a weak defense. The asking happened in the contract. The Authors Guild has pushed for exactly this kind of transparency, publishing model contract clauses that ask authors to disclose AI-generated text and that protect human authorship. If your manuscript agreement is vague, the move is simple: ask your editor what counts and what they expect you to declare. Better an awkward email now than a voided contract later.

Academic and hybrid presses: disclosure as the default

Academic and scholarly publishing leans hardest toward disclosure. Many journals and university presses now require authors to state any use of generative AI in a methods or acknowledgments section, and they’re firm that AI can’t be listed as an author because it can’t be held responsible for the work. That stance lines up with the U.S. Copyright Office, which has said purely AI-generated material isn’t protectable as human authorship. Hybrid and “author services” presses vary wildly, so read their terms the way you’d read KDP’s: looking for the generated-versus-assisted line and any required statement.

The throughline across all three lanes is inconsistency, not a universal rule. Each gatekeeper sets its own bar, and your job is to find that bar before you publish, not after.

A quick comparison

• Amazon KDP: Explicit declaration required for AI-*generated* text, images, and translations during upload, and editing the output doesn’t move it out of that category. Assisted use isn’t asked about. The declaration is private and doesn’t block publication by itself.
• Traditional trade publishers: Usually no public form, but AI clauses increasingly appear in contracts. You may be obligated to disclose or limit AI-generated material by agreement.
• Academic/scholarly: Disclosure of generative AI use is commonly required, often in a stated section; AI can’t be credited as an author.
• Hybrid/self-serve platforms: Varies. Treat the terms of service as the rulebook and look for the generated/assisted distinction.

For more on how human input shapes both authorship and how writing reads, it’s worth a browse the rest of our writing.

Frequently asked questions

Do authors legally have to disclose AI use in books? There’s no general U.S. law forcing authors to disclose AI use in a book. The obligations you’ll actually run into come from the platform or publisher, not a statute. Amazon KDP requires you to declare AI-generated content during upload as a condition of using the service, and a traditional publisher may write disclosure into your contract. Those are private rules you agree to, and breaking them can get a title removed or a contract voided.

What’s the difference between AI-generated and AI-assisted, and why does it matter? AI-generated means an AI tool created the content, and KDP says that label holds even if you edited it substantially. AI-assisted means you created the content and used AI to edit, refine, or brainstorm. KDP asks you to declare the generated category but not the assisted one, so the line decides whether you owe a disclosure at all.

Will disclosing AI use stop my book from being published? Usually no. On KDP, declaring AI-generated content is part of the normal flow and is informational. What gets books pulled is volume abuse and policy violations, not honest disclosure. Hiding content you were asked to declare is the riskier move.

Do traditional publishers require AI disclosure? Few have a public form, but many now include contract clauses asking you to represent your authorship and disclose or limit AI-generated material. The Authors Guild publishes model clauses to this effect. Read the manuscript agreement, and ask your editor if it’s unclear.

If I heavily edit AI output, do I still have to disclose it? Under KDP’s own framing, no. Content a model created counts as AI-generated even after substantial edits, so you’d still declare it. AI-assisted is for work you authored yourself with AI helping you refine. The honest test is who created the words. When in doubt, keep your drafts, and follow a publisher’s own definitions over KDP’s.

The bottom line

Disclosure isn’t one rule, it’s a question you answer differently at every door: an explicit KDP checkbox for generated content, a clause buried in a trade contract, a required statement in an academic acknowledgments section. The smart play is to know which door you’re using before you knock, and to put enough genuine authorship into the work that the answer is honest and easy. If you’re starting from a rough AI draft, shape a raw draft into prose that’s genuinely yours so the human authorship is real, not just a label on a form.