Publishing field guide

Amazon KDP AI Disclosure Policy: Prepare to answer KDP questions about AI-generated content accurately

Understand the difference between AI-generated and AI-assisted material, keep production records and review the current KDP Content Guidelines before upload.

Reviewed by Stefan Mitrović, Founder of Automateed · Updated July 16, 2026

Create a free preview Explore publishing guides

60-second summary

Quick answer

Amazon KDP requires publishers to answer questions about AI use during upload, distinguishing AI-generated content (created by AI, even with edits) from AI-assisted content (created by you, refined with AI tools) — with definitions that Amazon updates in its Content Guidelines. The safe practice is procedural: keep records of how text, images and translations were produced, read the current guideline at upload time, and answer from the records.

Real product steps

How to stay disclosure-ready with Automateed

The platform cannot answer Amazon’s questions for you — but a clean production record makes the answers trivial. Build the record as you work.

Workflow map

The amazon kdp ai disclosure policy path inside one account

Know what each part of your project is

Your book has distinguishable layers: generated chapter drafts, your edits and rewrites, generated images, uploaded images, AI-drafted metadata. The project itself is the map of what came from where.

Keep the brief and your editing notes

Save the instruction you generated from and note the scope of your rewriting per chapter. This is the difference between guessing and knowing when a form asks “generated or assisted?”

Track images separately from text

Amazon’s questions cover images too. Generated cover backgrounds and interior art are recorded facts in your project; uploaded photos are yours. Keep the distinction explicit.

Read the current guideline at upload time

Definitions have changed before and can change again. Open the KDP Content Guidelines (linked in this guide’s sources) during upload, not from memory or a blog post.

Answer literally and truthfully

Match your production records to Amazon’s current definitions and answer exactly. Disclosure is not a penalty — misrepresentation is.

This diagram mirrors the product steps above so the guide remains usable even when the interface evolves.

01
Know what each part of your project is
Your book has distinguishable layers: generated chapter drafts, your edits and rewrites, generated images, uploaded images, AI-drafted metadata. The project itself is the map of what came from where.
02
Keep the brief and your editing notes
Save the instruction you generated from and note the scope of your rewriting per chapter. This is the difference between guessing and knowing when a form asks “generated or assisted?”
03
Track images separately from text
Amazon’s questions cover images too. Generated cover backgrounds and interior art are recorded facts in your project; uploaded photos are yours. Keep the distinction explicit.
04
Read the current guideline at upload time
Definitions have changed before and can change again. Open the KDP Content Guidelines (linked in this guide’s sources) during upload, not from memory or a blog post.
05
Answer literally and truthfully
Match your production records to Amazon’s current definitions and answer exactly. Disclosure is not a penalty — misrepresentation is.

Every step above describes the current Automateed interface — open a free preview and follow along with your own project.

Create a free preview

The full guide

AI-generated vs AI-assisted: why the distinction matters

Amazon’s framework hinges on origination: content created by an AI tool is “AI-generated” even if you edited it afterward, while content you created and then refined with AI tools is “AI-assisted.” A book can contain both — generated first drafts you substantially rewrote, and your own chapters polished by tools. The classification question applies per content type (text, images, translations), which is why a layered record beats a single label.

The definitions above are a summary, not the policy. The current KDP Content Guidelines page is the only authoritative wording, and this page links it as a source.

Production records worth keeping for every book

Five artifacts cover nearly every disclosure and rights question: the generation brief, the outline as approved, notes on the depth of your editing per chapter, the origin of every image, and dates. In an integrated project most of this exists automatically — the project is the record. Export or note anything that lives outside it. Records also protect you beyond Amazon: rights disputes, store audits and even co-author disagreements resolve faster with a paper trail.

Disclosure and the market: what readers and stores actually punish

Stores and readers punish misrepresentation and low quality, not tools. A well-edited, accurately-described AI-assisted book faces no structural disadvantage; a mislabeled or unedited one accumulates returns and reviews that no disclosure setting can fix. Treat disclosure as one honest sentence in a trustworthy operation — and put the real effort into the editing passes that make the question boring.

Decisions that change the result

Build a disclosure-ready timeline (so you never “reconstruct” from memory)

When Amazon asks whether content was generated or assisted, the safest answers come from a timeline, not from recollection. A timeline means you can point to when a draft was produced, when you edited it, and what tool (if any) touched it. The tradeoff is effort: a complete record takes a little discipline during production, but it prevents slow, error-prone reconstruction later when the deadline is near.

A practical approach is to create one simple “production log” per book with entries that look like: date, content type (text chapter, back matter, cover art, interior illustrations, translation), source (tool name or “self-authored”), and your action (rewrote, summarized, refined prompts, retouched image). You do not need to expose prompts to Amazon; you only need an internal trace that lets you answer the upload questions consistently with the current KDP definitions.

How to classify mixed books without oversimplifying

Many KDP manuscripts are mixed even when the author feels they are “mostly theirs.” Classification gets tricky when you have multiple stages: for example, a chapter outline drafted with AI, followed by a human-written chapter, followed by AI polishing for style, and then a last manual pass for structure and facts. The decision you are making is not “Was AI involved at all?” but “Which content was created by AI, and which content did you create and then refine with AI tools?”

A simple rule that helps without guessing is to classify per artifact and per stage. If you start from an AI-produced text output and then substantially revise from that output, the result still reflects AI-generated material as described by Amazon’s framework. If you write the text yourself and then use AI to improve wording or consistency of your existing sentences, that tends to fit AI-assisted. For images, record whether the final visual was produced by a generator or uploaded/created by you, and be consistent about whether the input was your own reference or a generated prompt-based output. When in doubt, review the live KDP Content Guidelines at the moment you complete the upload questions and align your internal log to their wording.

Verification steps before you upload: reduce the chance of contradictory answers

Amazon forms can include multiple questions that appear to ask different versions of the same thing. Contradictions usually happen when answers are based on different assumptions: you might label one section as assisted but later label another as generated without a clear record. To avoid this, do a quick cross-check using your log: list every place AI was involved and confirm that each line maps to a single answer category that you can justify from your recorded origin.

Second, verify per content type. Text, images, and translations often have separate questions or separate wording that causes different interpretations. Keep a content-type checklist next to your draft so you do not carry a “text decision” over to “image decision.” Third, do a “definition check.” Open the KDP Content Guidelines and confirm the current definitions you are using. This is a decision point: treat the guideline as controlling even if it differs from what you remember from an earlier publication.

Worked example

Worked example: answering KDP AI questions from a mixed workflow

You are publishing a 120-page nonfiction ebook. You did not use AI to write the final narrative from scratch. However, you used AI to generate (1) a chapter-by-chapter outline, (2) draft examples to stimulate writing, and (3) a cover background image. You then wrote the chapters yourself, using the outline as a roadmap and discarding most generated examples. Finally, you used an AI tool to polish style and fix continuity after you wrote the chapters, and you created a final cover by combining the generated background with your own typography and a few custom elements.

01
Create a per-artifact log entry
Text outline: Created with AI tool (used as an outline reference). Chapters: Your original writing (based on your outline), then AI polishing for style and consistency. Examples: Some retained passages came from AI drafts; other examples were replaced or rewritten by you. Translation (if any): none for this example. Cover background: AI-generated. Final cover: composite created by you using the generated background plus your own typographic/layout work.
02
Decide classification per content type
For images: the cover background began as AI-generated output, so you record the cover’s underlying origin as AI-generated (even though you edited/composited it). For text: because you wrote the chapters yourself but also used AI to produce some draft material you incorporated, your log helps you answer precisely. If the upload form asks whether the book contains AI-generated content, you answer based on the parts that originated as AI output and were then used. If the form distinguishes between AI-generated and AI-assisted, you distinguish: AI-generated applies to any text you kept from AI drafts; AI-assisted applies to the parts you authored and then polished.
03
Cross-check form fields against the log
You review the upload questions and mark answers only where the log provides a direct mapping. If the form asks about images, your cover background is recorded as AI-generated. If it asks about text, you rely on the log lines that say which passages were retained from AI drafts and which chapters were your own writing polished with AI.
04
Align with the live KDP definitions at upload time
Before submitting, you open the KDP Content Guidelines and confirm the definitions you are using for terms like generated and assisted. If the wording suggests a stricter interpretation for “generated” (for example, AI-produced content that you integrated), you adjust your internal classification accordingly so the upload answers match the controlling policy text.

Your answers become straightforward lookups from your log: cover imagery aligns with the AI-generated origin; text answers split between AI-generated excerpts you incorporated and AI-assisted polishing you applied to your own writing. The main protection is not perfection—it’s consistency backed by an auditable timeline.

Avoidable mistakes

What usually breaks this workflow

Answering from a single label instead of per artifact

Many publishers remember “I used AI” and choose one broad option. But KDP questions can treat text and images separately, and mixed workflows produce mixed origins. If you collapse everything into one label, you can end up contradicting another question later in the same upload.

Using an old definition from memory or a past article

Even when your workflow stays the same, the definitions can change. If you do not re-check the live KDP Content Guidelines at upload time, you may answer with definitions that no longer match the current questions.

Confusing “AI helped edit” with “AI generated”

A common error is assuming that substantial human rewriting automatically changes generated content into assisted content. A record of origin matters. If the starting material for a passage came from AI output that you incorporated, you should treat that as AI-generated according to Amazon’s framework unless the live guideline wording indicates otherwise.

Not separating image production modes

Cover and interior images can be the largest source of confusion because authors often remember text but forget that backgrounds, textures, or illustration styles came from a generator. If you do not track image origins separately, you can accidentally answer “no” to an image-related question while your cover background was generated.

Keep the workflow moving

What to verify before acting on amazon kdp ai disclosure policy

Run these checks against the actual manuscript, files and reader journey before publishing.

Current Amazon wording is used

Text and images are both considered

Rights are documented

No disclosure is guessed from an old article

Continue the exact workflow

Tools and guides that belong after amazon kdp ai disclosure policy

Amazon Kdp Publishing Guide Self Publishing On Amazon Kdp Editorial Policy Editing Ai Books Ai Book Writing Kdp Formatting Images For Books State Of Ai Self Publishing

Editorial note

What this guide does and does not prove

This page is a practical workflow, not a promise of sales, ranking, publishing approval or a specific reader outcome. Platform rules and professional requirements should be checked at the point of use.

Questions specific to Amazon KDP AI Disclosure Policy

Before you start

Does Amazon reject AI-generated books?

Amazon accepts AI-involved books under its Content Guidelines, asks how content was produced, and holds all books to the same content and quality policies. Misrepresentation, not the tooling, is the risk.

Do buyers see my disclosure answers?

The questions are part of the publishing flow rather than a public badge on listings today — but policies evolve, which is exactly why answers should be accurate regardless of visibility.

Is heavily edited AI text still “AI-generated”?

Under Amazon’s current framework, content created by AI remains AI-generated even after editing. Substantial rewriting does not reclassify it — read the current guideline for the exact wording.

What about AI-generated covers and interior images?

Images are covered by the same questions. Record which visuals were generated and which uploaded; answer per the current definitions.

Does using AI for research or outlining count?

Assistance in ideation and refinement of your own writing falls on the AI-assisted side of the current framework — but verify against the live guideline at upload.

What records should I keep?

The brief, the approved outline, per-chapter editing notes, image origins and dates. Your Automateed project preserves most of this by existing.

Can my account be affected by wrong answers?

Policy violations — including misrepresentation — can affect listings and accounts under Amazon’s terms. Accurate answers cost nothing; inaccurate ones risk the catalog.

Do other stores have similar policies?

Several marketplaces now ask about AI involvement, each with its own definitions. The record-keeping habit satisfies all of them; the reading-the-current-policy habit must be repeated per store.

Where is the authoritative policy text?

Amazon’s KDP Content Guidelines page — linked in this guide’s official sources. Read it at upload time; summaries (including this one) age.

Does disclosure hurt sales?

Quality and accuracy drive sales outcomes. A disclosed, well-edited book with a clear promise outsells an undisclosed mess in every observable way that matters — reviews, returns, rank.

What should I do if I used an AI tool to translate or localize text, but later rewrote the translated passages myself?

Create log entries for the translation stage: record what was originally translated and whether you accepted AI-produced translation output or treated it only as a rough draft. Then compare your log to the live KDP Content Guidelines definitions. Even with later rewriting, the origin of the translated text you used matters for generated vs assisted classification.

If I generated a table, outline, or bullet list with AI and then rebuilt it manually, how do I classify that text?

Track the boundary between “used as reference” and “incorporated as generated content.” If you copied and retained AI-produced wording, that portion remains tied to AI output. If you used the AI output only for structure or ideas and then wrote the list from scratch with your own wording, your log may support an assisted classification for the final text—verify against the live guideline wording and answer consistently per artifact.

Explore next

Use your own topic

Test the “Read the current KDP guideline” direction with a free preview.

Review the outline, visual direction and available chapters before deciding whether to continue the full project.