What Are Amazon KDP’s AI Disclosure Rules for Authors in 2026?

KDP’s AI Disclosure Rules at a Glance

Amazon KDP requires authors to disclose AI-generated content when publishing or updating a book, primarily through the Kindle Direct Publishing (KDP) system. The disclosure applies to text, images, illustrations, and other material in a book that was published using the output of a generative AI system. It does not require authors to report every use of grammar checking, spelling correction, translation, brainstorming, or other AI-assisted tools. The key distinction is authorship responsibility: if a human creates, extensively edits, and verifies a manuscript, the book may be treated as human-created; if the author simply publishes largely generated material, disclosure is required.

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?

There is no universal numerical threshold in Amazon’s rule. Authors should not calculate a precise percentage and assume that crossing it makes disclosure optional. A book containing a small amount of original writing but many untouched AI-generated scenes may still be AI-generated, while a detailed human-written book edited line by line from a model’s output may fall on the assisted side of the policy. The publisher is responsible for making an honest determination and for keeping the disclosure accurate after later edits.

As of September 30, 2026, the practical advice for indie authors is straightforward: disclose AI-generated books during the publishing process, disclose material changes when updating a title, and document the human creative process if the work is close to the boundary. Amazon may remove, restrict, or investigate noncompliant books, and a disclosure can affect how readers understand the book even when no violation has occurred.

What Counts as AI-Generated Under KDP Policy?

KDP’s policy focuses on whether the published work is the output of a generative AI system. Examples commonly covered include a novel whose chapters were written mainly by a model, a nonfiction book assembled from generated explanations, or illustrated content in which an author supplies prompts and publishes the resulting images with little meaningful human revision. Disclose this material in the title setup or update process. The declaration is meant to tell readers and Amazon that the book was not created in the same way as a conventional author-written manuscript.

“AI-assisted” is a separate category. If an author uses a model to suggest alternate wording, identify typos, check grammar, generate a title, outline possible scenes, or improve a passage that the author then evaluates and rewrites, KDP generally does not require a disclosure. The distinction is not simply whether software was used. It is whether the work remains essentially generated output or whether the author has performed substantial creative authorship.

Amazon has not published a bright-line percentage for “substantial” human involvement. Authors should therefore avoid claims such as “70 percent human means no disclosure” unless a specific Amazon policy or written Amazon determination supports that conclusion. A safer test is to ask: Did the author write, select, organize, correct, verify, and meaningfully reshape the material? If the answer is mostly no, disclose it. This interpretation can depend on facts such as the number of model-produced passages, the quality of editing, whether the author can explain the finished book, and how much original material the author added.

The burden of truth is especially important where the book is informational. AI-generated advice about health, finance, law, or safety can be inaccurate, and disclosure does not transfer responsibility from the author. Authors remain accountable for permissions, factual accuracy, copyright, metadata, and the overall quality of the book. A disclosure explains the production method; it does not excuse a false or misleading claim.

How and Why the Disclosure Requirement Works

Amazon introduced the disclosure requirement as the quantity of synthetic material entering its self-publishing marketplace increased. The Guardian reported in 2023 that self-publishers must declare whether content sold on Amazon’s site is AI-generated, while GeekWire reported that Amazon’s policy distinguishes between “generated” and “assisted” content. The commercial rationale is trust: readers should know when they are buying a conventionally authored work, when a model produced a large portion of the text, and when human authorship still played a role.

The mechanism is built into KDP rather than left entirely to public labeling. Authors encounter the relevant declaration during title creation or an update workflow, and the answer becomes part of the information Amazon can review during catalog management. A book marked as AI-generated can be subjected to additional scrutiny, especially after complaints or automated detection signals. Amazon may ask for clarification, limit sales, remove the title, or take other action if it concludes that the disclosure was inaccurate.

The system is not a guarantee of either quality or misconduct. An author can disclose AI use and still deliver a carefully researched, genuinely useful book. Conversely, a book with no disclosure can be poor, misleading, or infringing. The label is a transparency mechanism, not a quality seal. Authors should avoid using “AI-generated” as a marketing substitute for transparent authorship information, and they should avoid withholding the declaration to make a book appear more traditionally written.

Amazon’s policy also sits within a wider environment of copyright, platform, and legal questions. Copyright treatment can differ by jurisdiction and by how much human authorship exists in particular portions of a work. KDP disclosure is a platform rule, not a general legal opinion about whether the entire book is copyrightable. Authors should separately review the rights to every image, model, source, quotation, and supplied training or reference material.

When Authors Must Disclose AI-Generated Material

An author generally needs to disclose AI-generated content during the initial KDP publishing process. The relevant point is not when the author first drafts the manuscript; it is when the work is prepared for publication and its production method must be represented. If the author starts with a model-generated novel, rewrites the text into a substantially different work, and then publishes the revised version, the disclosure decision should reflect the actual published manuscript, not merely the earlier draft.

The same principle applies to updates. A title originally based on a conventional manuscript does not automatically require a new disclosure merely because the author later uses AI to proofread it. If a later edition adds large amounts of generated text, new generated illustrations, or materially changed passages, however, the author should review whether the updated book is now AI-generated. Amazon may request changes to the disclosure when a new edition or significant content update alters the relevant facts.

Authors should disclose material that readers would reasonably expect an author to have created directly. This includes a full short story generated by a model, a nonfiction manuscript created mainly from model responses, translated or rewritten text that remains essentially machine-produced, and a collection in which a substantial share of the pages are generated. A human-written preface, title, or cover may not by itself change the status of a predominantly generated manuscript. Conversely, a single AI-assisted outline does not automatically transform an original book into generated content.

The safest approach is to make the disclosure before release and to correct it if later editing changes the answer. Waiting until after complaints or a platform review is not a sound policy strategy. Authors should preserve drafts, editing notes, research files, and version histories, especially when the book contains both human and AI-produced elements.

Begin by separating the book from the software used to make it. List the material that is conventionally human-created: original drafting, researched analysis, rewritten passages, selected sources, final editing, and human-designed visuals. Then list material produced by generative systems, including generated chapters, summaries, illustrations, or passages that were published with minimal changes. Do not treat a polished prompt as equal to an original manuscript; the published work is what matters.

Next, evaluate the extent of human authorship. Can the author explain the argument, verify every factual claim, recognize errors in the text, and explain why particular passages were retained or changed? For nonfiction, inspect every citation and statistic. For fiction, read the entire manuscript and revise voice, continuity, characterization, and plot. For image-rich books, check whether the images are original, licensed, publicly domain, or otherwise cleared for commercial use.

The workflow should end with a written decision before uploading. If the book is clearly generated, select the AI-generated disclosure. If the tool only helped with narrow tasks, document the assistance and proceed without a generated-content declaration. If the facts are uncertain, the prudent choice is to ask Amazon support or disclose and explain the circumstances rather than rely on a guessed percentage.

The decision should be revisited after major edits, but not treated as a permanent loophole. Do not publish generated text, add a cover, and leave the disclosure blank. Do not assume that a human name in the metadata changes the production method. Do not call a book “100 percent human” when the author merely corrected spelling in a model-written manuscript. Accurate records also help if Amazon asks about the title later.

Common Mistakes and Enforcement Risks

The most common mistake is treating AI-generated and AI-assisted as interchangeable terms. They are not. A grammar checker that flags a typo is different from a model that writes several chapters that the author publishes unchanged. Authors can also confuse a tool’s involvement with the final book’s status: using a large language model does not automatically make every sentence generated, and using a lesser-known model does not remove the disclosure obligation.

Another mistake is inventing a numerical safe harbor. KDP policy does not state a general 10-percent, 30-percent, 50-percent, or 70-percent threshold in the public guidance. Numbers circulating in blog posts may describe an author’s interpretation, not an Amazon rule. The better response is to examine the creative process and publish an honest answer. If an author cannot explain what was written, researched, verified, and revised, the risk is higher regardless of the percentage.

Authors should also avoid assuming that disclosure resolves legal or commercial problems. It does not guarantee copyright protection, remove the need for permissions, validate health or financial advice, or prevent reader complaints. A generated book can still violate KDP content guidelines, infringe someone else’s rights, contain misleading metadata, or fail quality standards. Conversely, authors should not use automation to mass-upload low-quality generated books and expect the disclosure to make the approach safe.

Amazon may respond to a failure to disclose through inquiry, metadata correction, sales restrictions, or removal, depending on the circumstances. The exact consequence is not always public and can depend on the title, complaint history, and policy findings. The appropriate response is to remain factual, correct inaccurate information, and avoid submitting duplicate or altered versions to evade review.

Cost, Alternatives, and Publishing Strategy

KDP’s standard publication model does not charge authors a separate fee for making the required AI disclosure. The relevant costs are the time needed to prepare, edit, verify, and disclose the book, plus any costs associated with using third-party AI or image tools. AI subscriptions can range from free browser-based plans to paid individual and business tiers, often marketed in monthly or annual billing. A subscription price is not the same as the cost of a book: authors may need to budget for research, cover design, editing, legal review, and proofing in addition to software.

Authors who want to avoid generated-content disclosure should use AI conservatively as an assistant, not as the principal author. That may mean using it for brainstorming, checking structure, suggesting headline alternatives, or catching repetitions while the author retains control of the manuscript. It is not a guarantee that Amazon will agree with every author’s interpretation, so the author should preserve evidence of substantial human writing and editing.

Direct publishing, assisted publishing, and a hybrid service are different alternatives. Assisted publishing can provide editors and production staff, but the author must still disclose AI-generated content accurately. A hybrid publisher may offer stronger editorial oversight while retaining a conventional publishing model, although terms, advances, royalties, and rights vary. Retaining an editor does not automatically turn generated prose into human-authored prose; the human creative contribution must actually shape the work.

A disclosed AI-assisted or AI-generated book can still be commercially viable if it offers genuine utility, transparent authorship, appropriate pricing, and a clear value proposition. Amazon does not state a universal sales threshold at which a book becomes AI-generated, and there is no general platform rule that every generated book must be priced at a specific amount. The sensible financial test is whether the expected royalties exceed editing, production, advertising, and platform-related costs.

What Authors Should Do Before Uploading a Title

The most defensible process is to inventory the manuscript before completing the KDP workflow. Identify all generative tools used, including writing assistants, image generators, translation systems, and automated research products. Mark the content that came directly from the tool, the content edited from it, and the content written independently. For a boundary case, ask a qualified editor or Amazon support for guidance rather than making a technical interpretation in good faith.

Authors should also review the book itself. Verify dates, names, quotations, legal claims, and statistics against reliable sources. Confirm image rights and disclose any third-party material in the appropriate metadata. If the book contains substantial generated content, make the declaration during the publishing or update step and ensure the final listing does not imply a level of traditional authorship that the book does not support.

On September 30, 2026, the safe answer is simple: disclose a book that is published as the output of generative AI, especially when model-produced text or images make up the central content. Do not disclose merely because you used a spellchecker or asked for a title suggestion, but document meaningful assistance and make a substantial human contribution. The policy is best understood as a transparency obligation, not as a numerical test. If the author is unsure, conservative disclosure and accurate explanation are more defensible than silence.