Amazon KDP’s AI Disclosure Requirement in 2026: What Self-Published Authors Must Report Before Publishing
A steady supply of articles describes KDP’s AI disclosure rule as a 2026 requirement. It is not new. Amazon announced it on September 7, 2023, following discussions with the Authors Guild, and the substance of the rule has not changed since. What has changed is how many authors now encounter it, because the share of manuscripts touching an AI tool at some stage has grown considerably.
That matters for a practical reason: authors searching for “the 2026 rules” often find summaries written to rank rather than to inform, and the distinction the policy actually turns on gets flattened. The requirement is narrow, it is answered at upload, and it hinges on a single definition that a lot of authors get backwards.
What the Policy Requires
Amazon’s KDP content guidelines state the obligation directly: “We require you to inform us of AI-generated content (text, images, or translations)” when publishing a new title or editing an existing one.
Three categories are in scope:
- Text — the manuscript itself
- Images — both cover art and interior artwork
- Translations — including a machine translation of a manuscript the author wrote themselves
The disclosure happens inside the publishing workflow. There is no separate form, no attachment, and no requirement to identify which tool was used or which portions of the book are affected. It is a question answered during upload.
The Line That Matters: Generated vs. Assisted
This is where authors most often misread the policy, and the guidelines are unusually clear about it.
AI-generated content is content an AI tool created — and it stays AI-generated no matter how much editing follows. The guideline language is explicit: “If you used an AI-based tool to create the actual content… it is considered ‘AI-generated,’ even if you applied substantial edits afterwards.”
AI-assisted content is content a human created, which an AI tool then helped refine. Brainstorming, outlining, grammar and error checking, suggested edits, tightening prose the author wrote — none of that triggers disclosure.
The pivot is authorship of the first draft, not the amount of human labor in the final one. An author who generates a chapter and then rewrites 80% of it has AI-generated content to disclose. An author who writes a chapter from scratch and runs it through an AI editing pass does not. Heavy revision is not a cure, and treating it as one is the most common compliance error.
The same test applies to artwork. A cover generated from a prompt is AI-generated even after extensive manual retouching. A photograph or illustration the author made, cleaned up with AI-powered tools, is assisted.
What the Disclosure Does Not Do
Two points are worth being precise about, because assumptions run in both directions.
It is not reader-facing. The disclosure goes to Amazon, not to the public. There is no badge, label, asterisk, or warning added to the product page. A disclosed title’s listing looks like any other title’s listing.
This is the specific gap the Authors Guild flagged when it called the policy a welcome first step. The Guild had asked for disclosures on book listings so readers could make informed choices, and noted that Amazon “has not yet agreed to add disclosures to all book listings for AI-generated books on its other channels outside of KDP.” Its stated concerns were AI-generated titles climbing bestseller lists and content farms appropriating the names and styles of known authors.
Disclosing does not, by itself, block publication. The policy is a reporting obligation, not a prohibition on AI-generated books. Answering honestly is what the rule asks for.
The Obligation That Outlives the Checkbox
The disclosure is the visible part. The heavier requirement sits next to it: publishers must verify that AI-generated content complies with intellectual property rights and the rest of Amazon’s content guidelines.
That obligation does not transfer to the tool vendor. An author who publishes AI-generated cover art has represented that the art does not infringe, and the practical difficulty is that generative models offer no provenance trail to support that representation. The same applies to text that may reproduce protected expression from training data. These questions sit on top of the unresolved authorship and ownership issues that determine whether AI-generated material attracts copyright protection at all — and separately, whether a work with no protectable authorship can support the exclusivity most publishing arrangements assume.
Authors relying on AI-generated components should also understand that a KDP account is governed by Amazon’s terms, and that account termination has historically been where content-guideline disputes actually land. Accurate disclosure is cheap; a misrepresentation discovered later is not.
Practical Reading
Answer the question at upload against the first-draft test, not against a sense of how much work you put in. Keep a record of which components of each title were generated and which were assisted, because you will be asked again on every future edit of that title, and reconstructing it from memory two years later is unreliable.
Where AI-generated images are involved, the disclosure is the smaller half of the problem. The representation that the material is clear of third-party rights is the part with consequences, and it is entirely yours.
This article is editorial and informational, not legal advice. Consult a licensed attorney about your specific situation.