What are Amazon KDP's AI disclosure rules for self-published books in 2026?

Amazon KDP requires authors to disclose AI-generated content when publishing through Kindle Direct Publishing. The policy, first introduced in September 2023 and refined in subsequent updates, draws a sharp line between AI-generated content (text, images, or translations created by an AI tool) and AI-assisted content (your own work that you created with AI help, such as brainstorming or editing). Only AI-generated content must be disclosed; AI-assisted content does not. This article explains exactly how the rules work, what you need to declare, where the disclosure appears, what happens if you get it wrong, and how the policy fits into the wider flood of machine-made books on the Kindle store.

The Direct Answer: What Amazon Requires

Also worth reading: How can self-published authors use KDP A+ Content to increase book sales and professionalism? · How many copies can a first-time self-published author realistically expect to sell? · What is the average number of books an author sells on Amazon?

When you publish a new title or edit an existing one through KDP, Amazon asks you to complete an AI content questionnaire during the book setup process. You are asked two questions: whether you used AI tools to generate text, images, or translations in your book, and whether you used AI tools to assist in creating it. If any part of your book was generated by AI — meaning the AI produced the actual words, artwork, or translation output — you must answer yes and specify which elements were generated.

The disclosure itself is not shown to readers on the product page. Amazon states it uses the information for internal purposes, including enforcing content policies and informing any future labeling decisions. This has frustrated transparency advocates who argue buyers deserve to know when a book was machine-generated, but as of August 2026, no reader-facing "AI-generated" badge exists on Amazon product pages. The obligation sits entirely on the author at upload time, and Amazon reserves the right to act against accounts that misrepresent their answers.

It is worth being precise about scope: the questionnaire covers text, illustrations/cover art, and translations. If you used ChatGPT to outline your nonfiction book but wrote every sentence yourself, that is assistance, not generation. If you prompted Midjourney to create your cover image, that is generation and must be declared. If you used DeepL or another machine translation service to produce a foreign-language edition, that counts as AI-generated translation under Amazon's definition.

Why Amazon Introduced the Rules

The policy emerged from a genuine operational crisis. By mid-2023, the Kindle store was being flooded with low-quality, rapidly produced AI books — some of them dangerous, such as mushroom-foraging guides generated without expert review. Reports circulated of hundreds of new titles per day coming from small numbers of accounts, and established authors found their names attached to AI-written books they never wrote, published by bad actors exploiting KDP's open model. Jane Friedman publicly documented scammers using her name on AI-generated titles, which forced Amazon to respond on brand-protection grounds as much as ethical ones.

Amazon's stated goals were threefold: protect the store's quality and customer trust, limit the volume of spammy uploads, and gather data on how prevalent AI content actually was. Publishers Weekly reported the guidelines when they launched, noting Amazon was among the first major retailers to formalize AI disclosure requirements rather than leaving it to copyright law, which in the United States remains unsettled for purely machine-generated works. The U.S. Copyright Office has repeatedly stated that material generated entirely by AI without meaningful human authorship cannot be copyrighted, which creates real commercial risk for authors who publish fully generated books — anyone could legally copy them.

There is also a less flattering motivation critics point to: data. Amazon is a major AI developer through AWS, and some observers have speculated that disclosure data helps Amazon understand and potentially benefit from the volume of AI-produced text flowing through its platform. No evidence confirms KDP disclosures feed training datasets, and Amazon denies this, but the concern recurs in coverage from outlets like ADTmag and GeekWire and is worth knowing about as context for why some authors distrust the questionnaire.

Generated vs. Assisted: The Line That Matters

Amazon's definitions are the heart of the policy, and getting them right determines whether you answer yes or no. AI-generated content means content created by an AI-based tool where the output is the substance of the work — the AI wrote the paragraphs, drew the image, or produced the translation, even if you then edited it. AI-assisted content means you created the work yourself and used AI only as a tool: outlining, research summaries, grammar checking, feedback on drafts, or generative tools used for ideation.

The gray zone is heavy editing. If an AI writes a draft and you rewrite 60 percent of it, Amazon's guidance treats the underlying generation as disclosable because the AI produced the original output. A practical rule many publishing consultants use: if you could delete the AI's contribution and still have your book, it was assistance; if removing it would gut the manuscript, it was generation. When genuinely unsure, disclosing costs you nothing visible to readers, while failing to disclose can cost you the account.

How to Complete the Disclosure: Practical Steps

The process takes under two minutes per title. During book setup in your KDP dashboard, after entering your title, description, and keywords, you reach the AI questions section before pricing. Answer honestly about each category: text, images, and translations. If you select yes for any category, follow-up fields ask you to describe which parts were generated — for example, "cover illustration" or "Spanish translation." Your answers save with the project and can be updated later if your workflow changes, such as when you commission a human illustrator to replace an AI cover.

Keep records of your own process. Save your drafts, prompts, and editing history so you can substantiate an "assisted" classification if Amazon ever queries it. Authors publishing series should note that each new title triggers the questionnaire again; answering consistently across a series matters, because contradictory answers across related books look like misrepresentation. If you work with ghostwriters, co-authors, or freelance formatters, ask them directly about their AI use — you are responsible for the accuracy of the declaration regardless of who did the production work.

For backlist titles published before September 2023, Amazon did not require retroactive declarations, and as of 2026 there is no blanket mandate to update old listings. However, if you republish, substantially revise, or move a title between formats, treat it as a new submission and complete the questionnaire. Some cautious publishers have voluntarily audited their catalogs, particularly where covers were AI-generated, to keep their accounts clean ahead of any future enforcement sweep.

Enforcement, Penalties, and What Actually Happens

Amazon has been deliberately vague about penalties, which frustrates publishers who want clear rules. The company states that failure to disclose may result in removal of the book or account-level consequences, but it has not published a strike system, fine schedule, or appeal process specific to AI misdeclaration. In practice, most observed enforcement has targeted the spam end of the spectrum: accounts uploading dozens of obviously generated titles per day have faced suspensions under existing content guidelines around quality and manipulation, with AI nondisclosure cited alongside other violations.

The realistic risk profile differs sharply by publisher type. A hybrid author who uses AI for one cover image and forgets to tick a box faces low probability of detection and modest consequences if caught. A volume publisher running an AI-content catalog faces meaningful risk, because Amazon's detection systems flag velocity patterns, duplicate structures, and reviewer complaints. Third-party watchdogs and author communities also report suspicious titles to KDP support, so social detection matters as much as algorithmic detection. The Guardian and other outlets have documented reader backlash against discovered AI books, which often triggers Amazon review even when the disclosure question was answered correctly.

One more legal layer compounds the risk: copyright. Under current U.S. Copyright Office guidance, purely AI-generated text and images lack copyright protection. That means a fully generated book can be scraped, republished, or audiobook-narrated by competitors with no legal recourse. Disclosure to Amazon does not fix this — it merely complies with platform rules. Authors building long-term intellectual property should weigh whether AI generation undermines the very asset they are trying to own.

Common Mistakes Authors Make

The most frequent error is assuming AI-assisted means AI-generated and over-disclosing. Telling Amazon your hand-written novel was AI-generated because you used ProWritingAid creates a false record that could complicate future copyright claims or marketing claims of human authorship. Read the definitions carefully: editing tools, research aids, and brainstorming do not require disclosure.

The second common mistake is forgetting images. Authors diligently disclose AI-written text but overlook that their cover came from an AI generator, or that interior spot illustrations were produced by DALL-E or similar tools. Images count independently, and the questionnaire asks about them separately. Translation is the third blind spot: authors who hire a translator assume human work, but some freelancers quietly run manuscripts through machine translation before light editing. Ask your vendors directly and put AI-use clauses in your contracts.

Fourth, authors sometimes believe disclosure will hurt sales and therefore lie. As of August 2026, disclosed AI content carries no visible label on Amazon, so honest disclosure has essentially zero commercial downside today — while a lie, once exposed by readers or journalists, creates lasting reputational damage that no platform policy can undo. Fifth, some authors assume the rules apply only to ebooks. They apply across KDP formats, including paperback interiors and covers uploaded through the same account. Finally, do not confuse Amazon's policy with retailer-wide standards: Apple Books, Google Play Books, Barnes & Noble Press, and Draft2Digital each handle AI differently, and several introduced their own disclosure or labeling mechanisms after Amazon moved first. Publishing wide means tracking multiple rulebooks.

How KDP Compares With Other Platforms

Understanding the competitive landscape helps you plan. Amazon was first among major retailers but not alone for long, and approaches differ in both strictness and transparency.

The practical takeaway is that Amazon's regime is currently the most explicit, which paradoxically makes it easier to comply with than vaguer policies elsewhere. Authors distributing widely should build a simple internal log per title recording which AI tools touched which elements, so they can answer every platform's questions accurately without re-auditing each time.

When to Act and Where This Is Heading

If you are publishing today, act now: complete the questionnaire truthfully for every new upload, audit any titles you revised recently, and document your creative process. There is no deadline pressure for compliant authors, but there is momentum pressure on the policy itself. Expect tightening through 2026 and beyond: industry pressure for reader-facing labels continues, the EU AI Act's transparency provisions push platforms operating in Europe toward clearer disclosure, and litigation over AI training data keeps the topic politically hot. A visible "AI-generated" badge on Amazon within the next couple of years is a plausible scenario, and authors whose disclosures are accurate today will face zero disruption if that happens.

For authors weighing AI use strategically, the calculus is straightforward. Using AI for assistance — editing, research, marketing copy outside the book — carries no disclosure burden and no copyright risk. Using AI for generation trades speed against ownership and platform risk, and increasingly against discoverability, since retailers and readers alike are growing hostile to undifferentiated machine-made content. The authors thriving in this environment tend to use AI narrowly, disclose cleanly, and invest their saved time in the human elements — voice, expertise, story — that neither algorithms nor regulators can commoditize.

Cost-wise, compliance is free: the questionnaire adds no fees, and no paid services are required to file accurate disclosures. The only real investment is administrative discipline — a spreadsheet logging AI tool usage per title costs nothing and protects your account, your copyright position, and your reputation in a market where all three are increasingly scrutinized.