Amazon KDP's AI Disclosure Rules in 2026: What to Disclose and What Happens If You Don't

Enforcement escalated sharply in 2025–2026. The one distinction that decides everything — generated versus assisted — and the ladder that follows.

The disclosure question is about origin, not final polish.

That one distinction — generated versus assisted — decides everything. Here is how to draw it correctly in 2026, before you upload. The compliance question is one of the biggest problems AI authors face, and it is now actively enforced.

The line that decides everything: AI-assisted vs AI-generated

Amazon's rule turns on who did the creative work. AI-assisted means you wrote and AI helped: brainstorming ideas and titles, outlining, grammar and spelling checks, editing and fact-checking prose you authored. None of that requires disclosure.

AI-generated means AI produced the content based on your prompts: text you then accepted, cover art, interior images, translations. This is the definition authors get wrong most often. Editing AI output does not move it to the assisted category. If AI created the words that remain in the book, the book is AI-generated for disclosure purposes — the question is about origin, not final polish.

The "substantial modification" bar

The one honest escape hatch is substantial modification. If you rewrite AI output at the level of restructuring, rewriting sentences for voice, and replacing weak passages — not light polish — the content crosses back into human authorship. The bar is high on purpose. A lightly edited AI draft is still AI-generated; a book where you rewrote the material in your own voice is not.

Practical rule of thumb: if the final text reads like you wrote it, the workflow was assisted; if it reads like the model wrote it and you tidied it, disclose.

What triggers disclosure

The disclosure question is about origin, not final polish.

What you must disclose

Three categories, no exceptions at the margin: AI-generated text (even after heavy editing), AI-generated cover art and interior images, and AI-generated translations. You answer the AI-content question separately for each book, in the Content tab of the publishing flow. Amazon may ask you to describe how AI was used — a plain statement such as "AI drafted chapters, which I then rewrote and edited" is sufficient.

What you don't need to disclose

Brainstorming, titles, and outlines produced with AI help; grammar, spelling, and style suggestions on text you wrote; using AI to edit or fact-check your own writing. The boundary is authorship: your creative work stays yours, and assistance does not trigger disclosure.

What the disclosure actually does — and doesn't

The disclosure is internal to Amazon. It is not shown on your product page, and Amazon states it does not affect search ranking, visibility, royalties, or promotional eligibility. There is no upside to hiding AI use and real downside if you do — undisclosed AI-generated content is a policy violation even when the content itself would have been permitted with disclosure.

The enforcement ladder (2025–2026)

The first year of the disclosure rule (introduced in 2023) saw sporadic enforcement. That changed in 2025 and has continued into 2026. Amazon's systems scan writing patterns, metadata, and upload speed to flag undisclosed AI content, and human reviewers check flagged titles. The consequences escalate: book removed from sale without advance warning; pending royalties on that title withheld; account warnings that accumulate toward suspension; account termination for repeated violations, eliminating access to every title and all pending royalties. The provenance side of this — why proving how a book was made is now part of the job — is covered in why provenance became the new editing.

The three-books-per-day cap

Separate from disclosure, KDP caps new title submissions at three per day per account — a limit added to the generative-AI policy and confirmed to the Authors Guild by KDP (the number may be adjusted). Updates to existing books do not count; paperback and ebook versions of the same title count separately; the cap applies per account, not per pen name. Most authors never approach it, but it exists because bulk unedited AI publishing was flooding the catalog — the same flood documented in why AI is making book discovery harder.

What this means for serious AI-assisted authors

The policy is not anti-AI; it is anti-bulk-unedited. The 2026 enforcement targets are accounts publishing churn-and-burn volumes of lightly edited generated text — the pattern behind the flood that now makes up roughly a fifth of new releases. A serious AI-assisted workflow — draft, rewrite, edit, verify — sits comfortably on the right side of both the disclosure rule and the quality standards, and treating AI output as a first draft rather than a finished manuscript is also the only way the writing is any good. The income picture confirms it: the authors making real money with AI are the ones who edit heavily and treat catalogs like brands, not the ones who upload generated drafts by the dozen.

That is the workflow this site is built around: a story system that keeps the book consistent while you do the rewriting, and transparent provenance for how the book was made. Disclose accurately, edit substantially, and the disclosure question stops being a risk and becomes a formality.

The disclosure question is about origin, not final polish. Disclose accurately, edit substantially, and the question stops being a risk.

The policy rewards the same thing readers reward: a book that was actually written, not merely generated. Structure first, then write — with Pacegram.