What are the KDP AI disclosure requirements for self-publishers on Amazon?

The Direct Answer: What Amazon Requires

Since September 2023, Amazon's Kindle Direct Publishing (KDP) platform has required authors to disclose AI-generated content when publishing or republishing a book. The policy was introduced in response to a flood of AI-generated titles hitting the Kindle store, a discoverability crisis that trade press outlets like Publishers Weekly and The Bookseller documented extensively. When you set up a new title in KDP, you now encounter a mandatory question in the publishing workflow asking whether your content was AI-generated, AI-assisted, or neither. This is not optional marketing information; it is a compliance question tied to your account standing.

Also worth reading: How does the EU AI Act define high-risk AI systems and what are the compliance requirements for publishers? · What is an AI disclosure statement for publishers, and how do I write one that actually protects my business? · What are Amazon KDP's AI content disclosure rules and how do I comply?

The disclosure requirement covers three categories of content: the book text itself, illustrations and cover images, and translations. If you used an AI tool such as ChatGPT, Claude, Midjourney, DALL-E, or any generative model to create text, artwork, or translations that appear in your published book, you are expected to answer yes. Amazon then asks you to specify which sections contain AI-generated material so it can apply appropriate review. Failure to disclose can result in content removal, account suspension, or in repeated cases, account termination.

It is worth being precise about what this disclosure actually does and does not do. Amazon has stated publicly that it does not display AI status to readers on the product page, at least not as a general rule. The disclosure is primarily an internal data-gathering and enforcement mechanism. Critics, including detection startups quoted by WIRED, have argued that Amazon could use this data to flag AI books proactively but largely does not, leaving much of the burden on author honesty rather than systematic verification.

Generated vs. Assisted: The Distinction That Matters Most

The single most important concept in the entire KDP AI policy is Amazon's distinction between AI-generated and AI-assisted content. Getting this distinction right determines whether you must disclose anything at all.

AI-generated content means text, images, or translations created by an AI-based tool where the output is substantially the finished work. If you prompted ChatGPT to write a chapter and published that chapter with light editing, that chapter is AI-generated. If you used Midjourney to create your cover art, that image is AI-generated. These require disclosure.

AI-assisted content means you created the work yourself and used AI tools in a supporting capacity: brainstorming ideas, outlining, refining grammar, checking spelling, or generating concepts you then executed independently. Amazon explicitly states that AI-assisted work does NOT require disclosure. If you wrote every sentence yourself but used Grammarly or ChatGPT to polish prose, you fall into the assisted category and answer no to the disclosure question.

The gray zone sits between these poles. A draft heavily generated by AI and then substantially rewritten by a human occupies contested territory, and Amazon's guidance does not draw a bright line based on percentage of revision. As a practical matter, if a reasonable reader would say the AI produced the core creative substance, treat it as generated and disclose it.

Why Amazon Introduced the Policy: Context and Criticism

Understanding why the rule exists helps you predict how strictly it will be enforced. By mid-2023, the Kindle store was experiencing what Startup Fortune described as a flood of AI-generated titles, including low-content books, plagiarized compilations, and fabricated nonfiction. High-profile incidents included authors discovering their names attached to AI-written books they never authored, sold on Amazon under their own brand. The Authors Guild had criticized KDP and Kindle Unlimited as early as 2019, arguing the program's structure attracts scammers who exploit the royalty system, and generative AI dramatically lowered the cost of producing fraudulent volume.

Amazon's response, announced in September 2023 and covered by Publishers Weekly, GeekWire, ADTmag, and The Guardian, was deliberately narrow. Rather than banning AI content outright, Amazon chose a disclosure-plus-limits approach: require declaration of AI-generated material while capping the volume of AI-generated titles any single publisher could upload per day, reportedly limited to three titles daily. This cap directly targeted the mass-upload operations flooding categories with hundreds of books.

Criticism persists on both sides. Detection startups told WIRED that Amazon could flag AI books using existing tools but generally does not, meaning enforcement depends heavily on reader reports and manual review. Author advocates argue the policy is toothless because dishonest publishers simply lie during disclosure, while some AI-friendly publishers argue the rules are vague enough that honest authors fear misclassifying legitimate assisted workflows. Neither criticism is wrong; the policy is best understood as a first-generation framework that Amazon will likely tighten as detection technology matures.

Practical Steps: How to Complete Disclosure Correctly

When you publish through KDP, the AI questions appear during the book setup process, after you enter your manuscript and before publication approval. Here is how the workflow functions in practice.

First, honestly audit your production process before you reach the screen. List every tool you used: writing assistants, image generators, translation services, and even AI features embedded in software like Canva or Adobe products. Second, classify each contribution against the generated-versus-assisted definitions above. Third, when the KDP questionnaire appears, select yes if any text, image, or translation is AI-generated, then specify which elements. You will be asked whether AI was used for the text, for images, and for translation separately, so a book with a human-written manuscript but an AI-generated cover requires a partial yes covering images only.

Fourth, keep records. Save your prompts, drafts, and edit histories demonstrating your workflow. If Amazon ever questions a title, documentation showing substantial human authorship is your defense. Fifth, remember that republishing triggers the same requirements. If you update an existing book's metadata or re-upload files, you may be presented with the disclosure questions again, and your answers should reflect current reality.

A common practical question concerns AI narration for audiobooks. Virtual Voice and similar AI-narrated audiobooks are labeled differently, and Amazon currently labels AI-narrated audiobooks visibly on the product page, unlike ebooks. Do not conflate ebook disclosure rules with audiobook labeling rules; they operate under separate mechanisms.

Common Mistakes That Get Accounts Flagged

The most frequent error is assuming light editing converts generated content into assisted content. It does not. Running a fully AI-drafted chapter through a grammar checker or asking another AI to rewrite it still produces AI-generated output in Amazon's framework. Authors who rationalize this way risk takedown notices.

The second common mistake involves images. Many authors diligently answer no about text while forgetting that their Canva template, stock-art composite, or Midjourney cover contains AI-generated imagery. Amazon's question explicitly covers illustrations and cover art, and image-based violations are among the easiest for reviewers to spot visually.

Third, authors misunderstand the translation category. Using DeepL, Google Translate, or GPT-based services to produce a translated edition creates AI-generated translated content requiring disclosure, even if the underlying original was entirely human-written.

Fourth, some authors believe disclosure harms sales and therefore omit it strategically. Since Amazon does not generally show AI status to ebook buyers, there is little demonstrated ranking penalty for honest disclosure, while the penalty for discovered nondisclosure includes removal and potential account termination. The asymmetry strongly favors honesty.

Fifth, serial publishers sometimes assume the daily upload limit applies per book rather than per account. The reported limit of three AI-generated titles per day applies across your publishing operation, and attempts to circumvent it through multiple accounts violate KDP terms and commonly result in linked-account terminations.

Enforcement, Consequences, and Realistic Risk Assessment

Amazon's stated consequences for failing to disclose include removing the book from sale, suppressing future publishing privileges, and terminating accounts for repeat offenses. In practice, enforcement appears reactive rather than proactive. WIRED reporting indicates Amazon relies significantly on reader complaints and competitor reports rather than running AI detectors across the catalog at scale. Detection startups claim their tools could flag AI books reliably, yet Amazon has not integrated them systematically, partly because false positives against human authors create legal and reputational exposure.

This creates an uneven risk profile. A single honest author who accidentally misclassifies borderline assistance faces low probability of action. A high-volume publisher uploading dozens of formulaic AI books weekly faces high probability of eventual flags, especially in saturated niches where human authors report competitors. The realistic calculus: your exposure scales with volume, pattern similarity across titles, and niche competitiveness, not with raw AI usage alone.

There is also a longer-term consideration. As of 2026, regulatory pressure around AI labeling continues building in multiple jurisdictions, including EU transparency rules affecting digital content. Amazon's internal dataset of disclosed AI content positions it to add reader-facing labels quickly if regulation or market pressure demands it. Authors who disclosed honestly from the start face no adjustment; those who did not may find historical nondisclosure harder to explain retroactively.

Cost Considerations: What Compliance Actually Costs

Disclosure itself costs nothing; the questionnaire is free and takes under two minutes per title. The real cost considerations sit upstream and downstream. Upstream, if you want to stay firmly in the assisted category, investing in professional editing runs roughly $0.01 to $0.05 per word for developmental and copyediting combined, meaning $500 to $2,500 for a typical 50,000-word novel. Commissioning human cover design costs $100 to $800 depending on complexity. These expenses buy you clean classification and stronger copyright positioning, since purely AI-generated elements occupy uncertain copyright territory under current US Copyright Office guidance, which denies registration to material lacking human authorship.

Downstream, nondisclosure carries contingent costs: lost royalties from removed titles, wasted advertising spend on delisted books, and the opportunity cost of rebuilding a terminated account. For a publisher running fifty titles, a single enforcement action can erase thousands of dollars in monthly income. Weighing a few hundred dollars in editing against account-level downside makes the economic case fairly one-sided for anyone publishing at scale.

For hobbyists publishing one personal memoir with modest AI assistance in outlining, the compliance burden is trivially small. For commercial operators, treating disclosure accuracy as part of standard operating procedure, with documented workflows per title, is the cheapest insurance available.

When to Act and How the Rules May Evolve

If you are publishing today, act at the moment of upload: complete the questionnaire accurately before hitting publish, because the disclosure is baked into the setup flow and cannot be cleanly retrofitted later without contacting KDP support. If you have backlist titles published before September 2023 that contain AI-generated material, you are not automatically required to amend them, but updating them proactively, particularly if you plan to revise or re-upload anyway, reduces future friction.

Expect evolution. The policy has already shifted once, moving from initial ambiguity toward firmer limits on upload volume, and industry coverage from The Bookseller and GeekWire suggests ongoing refinement of the generated-versus-assisted boundary. Plausible near-term changes include reader-facing labels for disclosed AI content, integration of third-party detection tools, and stricter thresholds defining how much human revision is needed to reclassify generated drafts. Authors who maintain clean, documented workflows will absorb these changes easily; those operating at the edges of the rules will find each tightening more expensive.

The pragmatic posture for any serious independent publisher is straightforward: use AI freely where it genuinely assists your creativity, disclose honestly wherever AI produced substantive output, document your process, and avoid high-volume patterns that attract scrutiny regardless of content quality. The disclosure regime is not going away, and treating it as routine compliance rather than an obstacle is both simpler and safer than any alternative.