KDP AI disclosure checker: must you disclose AI content?

When you publish or republish through Kindle Direct Publishing, the title setup asks whether your book contains AI-generated content — text, images, or translations made by AI tools. It is a straight yes-or-no gate in the publishing workflow, and Amazon requires an honest answer every time you publish or update.

This is a disclosure to Amazon, not a warning label on your book. Amazon does not currently require an AI notice inside the manuscript or the product description — it wants an accurate report in its own system so it can apply the right review and catalog treatment.

The widget above mirrors that gate: three questions covering the three content types KDP names. If any answer is yes, your KDP answer is yes. That one-to-one mapping is deliberate — there is no scoring, no gray zone in the tool, because there is none in the form.

Any final text an AI tool produced that appears in your book: chapters, paragraphs, poems, dialogue, summaries. Any image an AI model created: cover art, illustrations, diagrams, interior artwork. Any portion translated by AI. If a reader can see it and a model made it, KDP calls it AI-generated.

The part authors miss: substantial editing does not convert generated content into assisted content. Amazon's rule is explicit — text counts as AI-generated even if you rewrote most of it afterwards. Rewriting a ChatGPT chapter in your own voice is good craft, but it is still a yes on the form.

Machine translation trips up the same way. Running your manuscript through an AI translator, then polishing the output, produces AI-generated translated content under KDP's definitions — not a human translation with tooling help.

The other side of the line: you wrote the book and AI only helped along the way. Proofreading, grammar correction, and copyediting suggestions need no disclosure. Brainstorming ideas, outlines, or titles that you then wrote yourself needs none. Refining or retouching an image you created yourself — noise removal, color correction — is assistance, not generation.

The test is authorship of the final artifact. Did you produce the words on the page, with AI acting as editor, critic, or sounding board? That is assisted. Did the model produce words or pixels that survived into the book? That is generated, and the amount of later editing doesn't change the answer.

Most EarnDraft-adjacent workflows land clearly on one side. Drafting chapters with AI puts the text on the generated side (disclose); using AI to brainstorm chapter ideas you wrote yourself stays on the assisted side (no disclosure). When a workflow mixes both, the generated part governs — one yes anywhere means yes overall.

Concrete cases beat abstract rules. Here are three typical independent-author workflows and the honest KDP answer for each — find the row closest to your process.

Notice the middle row: it is the trap. The author did real work — restructuring, rewriting, fact-checking — and still must answer yes, because the underlying text was model-generated. Effort doesn't reclassify content; origin does.

In KDP title setup you will find the AI-content question alongside the other publishing declarations. Answer yes or no first — that is the load-bearing click. If yes, tick each box that applies: text, images, translation. Match the boxes to the widget's verdict, not to your gut feeling about how much you edited.

Then keep a disclosure page in your front matter naming the AI assistance in plain words. Amazon doesn't currently mandate in-book wording, but a simple page — what was AI-generated, what was yours, that you reviewed the output — protects you with readers and gives you a paper trail if KDP ever asks for details.

Finally, re-confirm on every update. Republishing a new edition or uploading a revised manuscript re-presents the question, and the correct answer can change between editions if your process changed. Make it a release-checklist line, not a one-time memory.

Amazon may reject the book, pull an already-live title, or restrict the account while it investigates — and misdeclaration sits in the same trust bucket as metadata gaming. The enforcement details shift over time, which is exactly why the safe play never changes: answer truthfully and keep records of your process.

An honest yes costs you nothing in most categories. Disclosure is routine now; thousands of disclosed AI-assisted books publish every week. What costs authors is a no that later looks dishonest — a reader report, a stylistic fingerprint, a paper trail in a prompt log you forgot about.

Your responsibilities don't end at the checkbox either. Amazon holds you accountable for accuracy, originality, and rights in AI-generated content the same as any other content. Review every AI passage for errors, plagiarism, and misleading claims before it ships — disclosure reports the content, it doesn't excuse it.

EarnDraft books are AI-drafted by definition, so the honest answer for a standard EarnDraft book is yes — AI-generated text. We don't hide that; we operationalize it. Every finished book ships with a disclosure page already in the files, and your listing answers come pre-filled, so the KDP form reflects reality before you touch it.

That handling covers the standard path: AI-drafted chapters plus an EarnDraft cover. If you replace the cover with your own photography or add human-written chapters, your answers change — re-run the checker above against the final files, not against how the project started.

And the boundary holds on scope: disclosure handling is paperwork accuracy, not legal advice. For genuinely uncertain cases — licensed material in training-style outputs, co-authorship questions — a publishing-savvy review beats any checklist, including ours.

Keep a one-page production log per book: which tools you used, what they produced (draft text? cover art? translation?), and what you changed afterwards. Store it with the manuscript or alongside your publishing files — with dates and tool versions, since “AI assistance” means something different every year. If KDP ever questions a declaration, that page turns a stressful week into a five-minute reply. For the full checklist version, read our KDP AI disclosure guide alongside this checker.

Apply one honest standard everywhere. Other stores differ — some mirror KDP's question, others rely on general originality warranties — but KDP's rule is the strictest widely-enforced one, so clearing its bar usually clears the rest. Disclose to every store what you'd disclose to KDP and keep the same in-book page in every edition; consistency is cheaper than remembering which version went where.

When choosing where to sell, add each candidate's AI policy to the comparison next to fees and reach. A channel whose rules you can't confidently meet is a channel you shouldn't launch on — no royalty rate compensates for a takedown quarter.