KDP AI Disclosure Rules: The Direct Answer
Amazon KDP requires authors to disclose AI-generated content in books sold through Amazon when that content was not created primarily through a human author’s own written or typed input. The disclosure applies to AI-produced text, images, illustrations, and other material created or substantially modified by generative AI. It does not ordinarily cover spelling, grammar, or punctuation tools, research assistance, brainstorming, or translation that the author subsequently wrote and revised. Amazon introduced these rules in September 2023 after growing concern about rapidly increasing numbers of low-quality books on Kindle. As of September 2026, the practical obligation remains: identify material that meaningfully qualifies as AI-generated, make the disclosure through the KDP workflow, and keep records showing how each part of the book was produced. A book is not automatically required to carry a visible warning label. The principal requirement is that the author disclose qualifying material during publication; Amazon may separately review flagged books, request information, remove content, or delay distribution. Authors should therefore treat disclosure as a publishing-compliance question, not as a marketing decision.
Also worth reading: How Should Authors Handle AI Disclosure in Publishing Contracts and Reader Expectations in 2026? · What are the AI book cover disclosure requirements for 2026 and how do they affect self-published authors? · What is an AI editing contract disclosure clause and why do authors need it?
The rule is narrower than the claim that “all AI use must be disclosed.” Amazon distinguishes between AI-generated and AI-assisted work. AI-generated content means the model itself produced material without the author directly writing or typing it, while assisted work generally means a human used software to check spelling, grammar, readability, organization, or factual consistency. The distinction matters because hundreds of authors now run every manuscript through Grammarly, ProWritingAid, or a comparable service, and those uses normally fall into the assisted category. An author who writes original sentences but asks an image model to create a cover must disclose that AI-generated image. An author who uses AI to suggest chapter headings, however, is performing a different act from asking it to compose a chapter the author never supplied.
Why Amazon Created the Disclosure Requirement
Amazon’s change followed evidence that generative-AI books were entering Kindle faster than readers, reviewers, and booksellers could evaluate them. The market problem was not simply that authors used a new tool. It was that books containing millions of apparently repetitive sentences, automatically assembled articles, or mass-produced biographies could consume scarce discovery positions while offering little editorial judgment. Search and recommendation systems often cannot tell whether a competent human editor controls a manuscript or whether a system generated large sections without meaningful human authorship. This makes disclosure useful not only to Amazon but also to readers deciding whether a book deserves their time and money.
The September 2023 announcement was a significant policy intervention, but it did not establish a universal legal definition of AI writing or resolve intellectual-property questions. Copyright status can be separate from disclosure status: a work may need disclosure even when the author’s copyright position is uncertain, and copyright may not attach to every purely AI-produced component. Amazon’s rule was designed primarily to improve marketplace transparency and quality control. Authors should not interpret permission to use AI as permission to submit mass-produced or misleading books. KDP also prohibits content that is misleading, violates intellectual-property rights, or infringes another person’s rights, regardless of whether AI was disclosed.
The timing remains important because this is an active compliance area rather than a settled publishing statute. Policies, interfaces, and enforcement practices can change, and national rules may evolve at the same time. A publisher preparing a September 2026 release should open the current KDP content guidelines in the KDP Help Center, inspect the disclosure prompts in the title setup workflow, and retain evidence of the creation process. Reporting a 2023 rule as though Amazon issued permanent, detailed legislation in 2026 would overstate what the available guidance says.
What Counts as AI-Generated and What Does Not?
Text classification depends on the author’s real contribution, not on whether a tool was merely named in a software subscription. If an author enters detailed instructions, writes substantial passages, edits every paragraph, verifies facts, and treats AI as a drafting partner, the final work can be considered human-led even if some model-assisted text remains. If the author asks a model to write a scene, biography, outline section, or entire manuscript from a brief prompt and publishes the result with only minor cleanup, the qualifying output is AI-generated. Rewriting an AI passage in a word processor can reduce the generated portion, but it does not automatically transform the work into non-disclosable assisted content.
Images are classified more directly. A cover, interior illustration, decorative graphic, or other visual produced substantially by a generative-image system normally falls within the disclosure requirement. Merely resizing, retouching, adjusting brightness, or applying a conventional photographic filter usually presents a different case because the underlying creative image was not generated by AI. A composite is judged component by component: a photographed author portrait placed on a conventional layout is not automatically AI-generated, while a synthetic fantasy portrait created by an image model is. Authors who combine several tools should preserve drafts, prompt records, source-image files, and editing histories so they can explain what each tool did.
| Feature | Usually AI-generated | Usually AI-assisted |
|---|---|---|
| Manuscript text | Model writes scenes, prose, biographies, or chapters from prompts | Grammar, spelling, readability, and consistency checks on human-written text |
| Brainstorming | Model supplies ready-to-publish passages or extensive draft prose | Model proposes topics, questions, or rough ideas that the author develops |
| Images | Model creates the principal cover illustration or interior artwork | Basic resizing, cropping, color correction, or layout work on an existing image |
| Editing | Unreviewed generated text is pasted into the manuscript | Author writes, edits, compares, and verifies the suggested revisions |
| Disclosure | Required in the KDP content declaration when applicable | Normally not required under Amazon’s AI-assisted category |
The strongest trigger is a book that includes text or images the author did not themselves create through written or typed input. Examples include an entire novel generated from a character brief, hundreds of automatically produced poetry sections, or a cover whose principal artwork came from an image generator. Disclosure is also appropriate when generative AI rewrites or substitutes a meaningful portion of an existing work without human authorship of the resulting language or visual. Authors should err toward disclosure when they are uncertain, particularly because a short model-written subsection can be hard to isolate visually in a finished book.
The location of the AI work does not remove the obligation. A generated cover, a text section in a reference guide, an AI-produced image in a children’s book, and model-written material in an audio-related companion publication can each raise a separate question. A professionally designed cover that contains a small AI-assisted element is not necessarily the same as an entirely synthetic cover. The author should classify the content at the level at which AI materially performed the creation. If a qualifying element appears in the uploaded files, the author should answer the KDP disclosure prompt before publication and correct the answer if the manuscript changes materially.
Do not wait for an author to accuse another writer of “using AI.” Reporting can trigger a review, but it is a poor compliance strategy. Reviewers may look for repeated syntax, implausible facts, generic transitions, fabricated citations, inconsistent formatting, or suspiciously uniform voices, although no single pattern proves machine authorship. Amazon can examine metadata and submission history, but authors are in the best position to document their process. A simple folder containing outlines, drafts, research notes, prompts, revisions, and source files can be more useful than a retrospective claim that all writing was “human-assisted.”
The KDP Disclosure Process and Recordkeeping
The process begins before publication, during preparation of the Kindle eBook details in KDP. The author should review the title and content prompts for questions concerning AI-generated text, images, or other material, and provide an accurate declaration whenever a qualifying answer is required. The precise screen wording may be updated over time, so authors should not rely on a screenshot from 2023. They should also reread the current KDP content guidelines rather than assume that a third-party blog’s interpretation controls. KDP’s official help material remains the primary operational reference for Amazon submissions.
Disclosure does not mean adding “This book was written by AI” to the cover unless Amazon separately requires or authorizes such a label. Nor does it permit the author to misstate the book as entirely human-written in the metadata. A transparent project note can be useful when the AI contribution is material, but the marketplace declaration and the public description should be consistent. If the book’s core proposition is AI-assisted creation, vague wording that implies traditional authorship can mislead readers independently of Amazon’s formal requirement.
Recordkeeping should cover both generated and human-created material. Useful evidence includes dated outlines, typed prompts, model responses, first drafts, source notes, fact-check sheets, image prompts, generated source files, and final revision histories. Authors publishing with a team should agree which person owns the declaration. A cover designer may be outside KDP and need to report whether a source visual was generated, while a coauthor may have used AI for a chapter the lead author never reviewed. The rights holder should collect declarations from every contributor and compare them with the uploaded book before approving release.
Costs, Platforms, and Alternatives to Undisclosed AI Production
AI disclosure itself is generally not presented as a paid KDP listing feature. KDP’s standard eBook and paperback publishing options are ordinarily free to use, although authors buy the printing cost for paperbacks, purchase proof copies, and may pay for editing, artwork, software, or marketing. Amazon can change fees or offer promotions, so authors should calculate the current terms in KDP rather than quote a timeless price. An author using a commercial language or image tool may pay from nothing for a limited plan to several hundred dollars annually for higher usage, but those subscription prices are vendor-specific and frequently change.
| Publishing route | Typical cost structure | AI disclosure issue |
|---|---|---|
| Human-written, AI-assisted edition | Free KDP setup; optional grammar, editing, and design costs | Assisted tools ordinarily do not require disclosure |
| AI-assisted cover made from licensed human artwork | Free KDP setup; optional design or stock-license fees | Disclose when generative AI materially created the visual |
| Human-authored book with a disclosed AI-written section | Free KDP setup plus any editing needed to review the section | Disclose the qualifying section accurately |
| Human-reviewed rewrite of model-generated drafts | Free KDP setup plus substantial editing and comparison costs | Human control does not automatically erase the original generated material |
| Fully generated book submitted as ordinary human work | Potentially free to upload | Misrepresentation creates review, removal, account, and trust risks |
Common Mistakes That Lead to Rejections or Retractions
The most common error is treating disclosure as a label for an entire book when the rule concerns the nature of particular content. Saying nothing is safe only when no qualifying generated text or imagery appears. The opposite error is assuming that one grammar check makes every use fully non-disclosable. Authors also confuse a tool’s output with their own: accepting a large block of model-written prose because it is “just a first draft” does not turn that passage into human-written text. The final publication history matters.
Another mistake is trusting a universal “AI detector.” Commercial detectors produce false positives and false negatives, especially for non-native English writers, edited text, formulaic genres, and short samples. Authors should not try to defeat detection; they should make an accurate disclosure and retain evidence of their process. Fabricated references, invented quotations, inaccurate biographies, and fabricated author experiences are separate quality concerns that disclosure does not legalize. Readers can be harmed by a confidently written but factually false book even if its manuscript is fully human-written.
A final error is treating a successful upload as proof of compliance. KDP can review a title after publication, and distribution can stop while Amazon examines a complaint or quality concern. Repeated undisclosed material can lead to a takedown or broader account enforcement, depending on the circumstances and policy language. The safest operating rule is simple: when material was generated rather than merely checked or formatted, disclose it; when the facts are uncertain, ask KDP Support in writing and preserve the response.
What Authors Should Do Before a September 2026 Release
Authors should classify the project before completing final production. Make a table of chapters, passages, images, covers, and supplementary materials, then mark each item as human-created, AI-generated, or assisted. Review the classification with every coauthor, editor, cover designer, and rights holder. Next, open the live KDP Help Center and compare the project with Amazon’s current AI-content language, because the September 2023 launch date and a September 2026 publication date do not guarantee identical interfaces or enforcement details. Finally, complete the metadata declaration accurately and keep the supporting files until the title has been stable in the marketplace.
The book should also be checked for ordinary publishing failures that AI may increase. Verify names, dates, quotations, citations, translations, page numbers, image rights, and the existence of organizations and sources. Have a human editor compare every final sentence against the draft and sources rather than accepting polished model prose because it sounds authoritative. For fiction, examine continuity, character voices, and repeated descriptions. For nonfiction, test citations manually and remove any reference the author cannot locate. A disclosure answers an origin question; it does not replace editorial responsibility.
Authors should act before publication, not after a complaint. A prepublication correction is usually easier than replacing a title, notifying buyers, withdrawing an audiobook, or responding to an intellectual-property claim. If a significant section is AI-generated, the author can disclose it, obtain a human rewrite, or remove it, but should not leave inaccurate metadata in place. For professional books with legal, medical, financial, or factual stakes, use qualified human review even when every sentence was generated. The best disclosure approach is neither secrecy nor exaggerated fear; it is a documented production process, an accurate marketplace answer, and editorial control over what readers receive.
The Practical Standard for KDP Compliance
As of September 2026, KDP’s core standard is content-based: disclose AI-generated text, images, or other material, while ordinary spelling, grammar, and similar assistance generally falls into the separate assisted category. The requirement is not triggered merely because an author owns software with AI features, and it is not satisfied merely because a model’s output was copied into a draft. The decisive question is who created the material through written or typed human input and who performed the creative work.
For an AI Publishing Consultant, the advice should be documented rather than alarmist. Amazon’s intervention addresses a real flood of low-control material, but blanket claims that AI is automatically “banned” from Amazon are inaccurate. Authors may publish books involving generative AI when they follow the applicable declaration, marketplace, quality, and intellectual-property rules. Books designed to flood KDP may still be removed even if fully generated and accurately disclosed, because volume, misleading conduct, copyright violations, or poor quality can trigger separate problems.
The defensible workflow is to inventory the AI contribution, edit and fact-check the manuscript, disclose qualifying material in the live KDP workflow, and retain provenance records. Authors who did that work already have a stronger defense than those who try to rely on detector scores or public accusations. Because policies can change after September 2026, the current KDP Help Center and direct support response should control when they conflict with an older article, including this one.