Amazon KDP requires authors to disclose AI-generated content when publishing through Kindle Direct Publishing. Since September 2023, when Amazon first introduced the policy, every author completing the KDP publishing workflow has been asked two questions: whether the text was generated by AI, and whether images were generated by AI. The disclosure is made during the book setup process and applies to new titles and any re-published or significantly edited editions. As of 2026, the rules remain in force, though enforcement remains one of the weakest points of the entire system — something every self-publisher should understand before deciding how honestly (or strategically) to fill out that form.
The Direct Answer: What You Must Disclose
Also worth reading: What is an AI disclosure statement for publishers, and how do I write one that actually protects my business? · Where can unpublished authors find publishers that accept self-publishing for novels? · How many self published authors are on Amazon?
Amazon's policy draws a sharp line between two categories of content. "AI-generated" content means text, images, or translations created by an AI-based tool, even if you substantially edited it afterward. If ChatGPT wrote your first draft and you rewrote half of it, Amazon still considers the book AI-generated. "AI-assisted" content, by contrast, means you created the work yourself but used AI tools to refine, edit, check grammar, brainstorm ideas, or generate concepts. AI-assisted work does not require disclosure.
This distinction matters enormously in practice. A novelist who uses ProWritingAid or Grammarly to polish their own prose discloses nothing. An author who prompts Claude to write chapters that they then lightly edit must select "yes" on the AI-generated question. Amazon has stated that when you disclose AI-generated content, you confirm the content complies with KDP's content guidelines, and you remain responsible for verifying quality, originality, and accuracy regardless of how the material was produced.
The disclosure itself does not appear on the public product page. Readers cannot currently see which books were flagged as AI-generated. This has been a persistent criticism from human authors who argue the market is being flooded with machine-written books with no way for buyers to filter them out. Startup Fortune and other outlets have documented the discoverability crisis this creates, particularly in low-content niches like children's books, journals, and genre romance where AI output volume exploded after late 2022.
Why Amazon Introduced the Rules
Amazon faced a genuine operational crisis. By early 2023, reports emerged of hundreds of AI-generated ebooks appearing daily, some containing fabricated information — including at least one high-profile case of a guide that suggested poisonous mushrooms could be eaten safely. Amazon capped self-publishing at three books per day per account in response, a limit that persists today and exists specifically because of AI volume abuse.
The company also needed legal cover. Under the DMCA and its own terms of service, Amazon needs a basis to remove infringing or problematic content, and AI-generated works frequently reproduce copyrighted material from training data without permission. By requiring disclosure, Amazon shifts liability toward the author: if you declare AI-generated content, you affirmatively warrant that it doesn't violate anyone's rights. If you lie on the disclosure and get caught, Amazon can terminate your account and withhold royalties.
There is also a data angle worth noting critically. Several commentators, including coverage picked up by WIRED and ADTmag, have raised concerns about what happens to disclosed AI-generated content — specifically whether Amazon might use submitted manuscripts to train its own generative models, which require vast quantities of text. Amazon has not offered authors a clear opt-out mechanism for training use of uploaded content, and skeptics note the disclosure system conveniently gives Amazon labeled data distinguishing human-written from machine-written text. Whether this concern is fully justified or speculative, it is a legitimate consideration for authors weighing where to publish.
How Disclosure Works Step by Step
When you set up a new ebook or paperback in KDP, the AI questions appear in the content section before you can publish. For text, you're asked: "Did you use AI tools in creating texts?" with options indicating whether the text was AI-generated, AI-assisted, or neither. The same structure repeats for images and for translations. Answering honestly takes under a minute; answering dishonestly takes the same amount of time but carries account-level risk.
For translations, the rule is notable: if you used an AI translation tool such as DeepL or Google Translate to produce a foreign-language edition, that counts as AI-generated content and must be disclosed. Many multilingual publishers were caught off guard by this, since translation had long been a standard localization practice. Amazon treats machine-translated editions the same as machine-written originals.
If you update a previously published book, you may be prompted to complete the AI questions retroactively when uploading revised files. Books published before September 2023 generally don't require retroactive disclosure unless you republish or make substantive changes. Amazon has not announced any program requiring authors to audit backlist titles.
Comparison: AI-Generated vs. AI-Assisted Content
| Feature | AI-Generated Content | AI-Assisted Content |
|---|---|---|
| Definition | Text/images/translations created by AI tools | Human-created work refined with AI tools |
| Disclosure required | Yes, mandatory during KDP setup | No |
| Examples | ChatGPT-drafted novel, Midjourney cover art, DeepL translation | Grammarly edits, manual research with AI chatbot help, brainstorming |
| Public visibility | Not shown to readers (as of Aug 2026) | N/A |
| Account risk if undisclosed | Termination, royalty withholding possible | None |
| Volume limits affected | Subject to 3-books-per-day cap | Same cap applies account-wide |
| Responsibility for accuracy | Author bears full responsibility | Author bears full responsibility |
Enforcement Reality: Weak Detection, Real Consequences
Here is the uncomfortable truth: Amazon's enforcement is largely reactive. WIRED reported that AI detection startups claimed Amazon could flag AI books at scale, but evidence suggests it largely doesn't proactively scan submissions. Detection of undisclosed AI content typically happens through reader complaints, copyright infringement claims from authors whose work was scraped into outputs, or manual review of suspiciously prolific accounts. Some AI-generated books have been removed, and some accounts terminated, but the vast majority of undisclosed AI content almost certainly passes through unnoticed.
That said, the consequences when caught are severe. Amazon can terminate your KDP account permanently, withhold unpaid royalties, and block you from future publishing. Because Amazon dominates roughly 65-70% of the US ebook market, losing KDP access is commercially devastating for most independent authors. There is no meaningful appeals process that reliably restores terminated accounts.
Critics also point out that the disclosure system creates asymmetric burdens. Honest authors spend time filling out forms while bad-faith publishers simply click "no." Meanwhile, legitimate hybrid workflows — where a human writes extensively with AI research assistance — face ambiguity that Amazon has never fully clarified. Publishers Weekly and GeekWire both noted confusion among authors about where editing ends and generation begins, and Amazon's published guidance has not materially resolved this since introduction.
Common Mistakes Authors Make
The first mistake is assuming covers are exempt. AI image generators like Midjourney produce most new indie covers now, and those count as AI-generated images requiring disclosure even when the text is entirely human-written. The second mistake is treating translation tools as neutral utilities rather than AI generation — machine-translated editions must be declared. Third, some authors believe heavy editing converts generated text into assisted text; under Amazon's stated definition, it does not. Editing AI output changes the degree of generation, not the fact of it.
Fourth, authors sometimes assume disclosure harms sales. There is currently no evidence readers can see disclosures, so honest answers carry no demonstrated ranking penalty. Fifth, multi-author operations often fail to coordinate: if a ghostwriter or co-author used AI, the publisher of record is still responsible for accurate disclosure. Finally, some authors skip disclosure on low-content books thinking nobody checks — but low-content categories are precisely where Amazon has applied the most scrutiny due to sheer volume abuse.
When to Act and What It Costs
Disclosure costs nothing and takes less than a minute per title during setup. If you published before September 2023 and haven't touched the files, no action is required. If you're republishing, updating, or launching anything new in 2026, answer the questions accurately at upload time. There is no fee, no certification process, and no third-party verification involved.
Timing matters more for strategic reasons than compliance ones. If you plan a large catalog launch, remember the three-per-day upload cap and build buffer time. If your workflow involves substantial AI drafting, consider investing in genuine human editing and originality checks before publication — both because Amazon holds you responsible for quality and because plagiarism claims against AI output are increasingly common as original authors recognize their work regurgitated in machine-written books.
Alternatives and the Broader Context
Other platforms handle AI differently. Draft2Digital asks about AI content in its terms but with lighter procedural requirements. Apple Books prohibits purely AI-generated works in certain categories. Smashwords (now merged with Draft2Digital) requires disclosure of AI-generated content. IngramSpark has its own content policies that prohibit mass-produced AI spam. None of these alternatives match Amazon's reach, so most authors comply with KDP's rules while diversifying distribution.
The regulatory environment may tighten further. The EU AI Act's transparency provisions and various US state proposals could eventually force platform-level labeling visible to consumers, which would change the calculus around disclosure considerably. Jane Friedman's ongoing FAQ coverage tracks these developments closely and remains one of the most reliable references for working authors navigating the shifting rules.
Bottom Line
Amazon KDP's AI disclosure rules ask a simple question and expect a truthful answer: did a machine generate this text, imagery, or translation? Declare AI-generated content during setup, keep AI-assisted work undisclosed, stay under the three-book daily limit, and never assume weak enforcement means zero risk. The system is imperfect, inconsistently enforced, and criticized from multiple directions — by human authors wanting visible labels, by AI publishers wanting clarity, and by privacy advocates wary of Amazon's data uses. But it is the rulebook for the largest ebook marketplace on earth, and ignoring it puts your account and royalties at unnecessary risk.