KDP AI Content Disclosure Checklist: Generated vs Assisted + Free Log
Does Amazon KDP require you to disclose a book written with AI? The answer depends on what the tool created, not just how much time you spent editing it. Amazon distinguishes AI-generated text, images, and translations from AI-assisted work. That distinction is easy to miss when one person researches, another drafts, a model rewrites, and a designer creates the cover.
This KDP AI content disclosure checklist turns the published rule into an auditable pre-upload workflow. It includes a free decision-log CSV, six realistic but fictional examples, and a worked inventory for a mixed-origin nonfiction book. The goal is to give an author accurate information for KDP's own disclosure process, not to provide legal advice or promise that any book will be approved.
Free KDP AI content disclosure checklist and decision log
Download the editable KDP AI content decision log (CSV). The worksheet has fourteen columns, six explicitly fictional examples, and eight empty rows. Record one row for each distinct source of text, image, or translation: a manuscript section, cover, illustration, or translated chapter. Keep the file in your own project; it is not an Amazon upload form and is not sent to KDP by Scribora.
Need to turn permitted research into a nonfiction draft first? Scribora provides a source-first workspace for an editable outline, chapter drafts, review, and PDF or EPUB export. Explore the free outline and sample-chapter workflow. You remain responsible for documenting the origin of generated content, checking rights and accuracy, and making the appropriate disclosures when publishing.
The KDP rule: generated versus assisted
The primary source is Amazon's KDP Content Guidelines, in the section titled “Artificial intelligence (AI) content (text, images, or translations).” Checked on November 17, 2026, the rule requires a publisher to inform KDP of AI-generated content when publishing a new book or editing and republishing an existing book. This includes generated cover or interior artwork, not just prose.
Amazon defines AI-generated as actual text, images, or translations created by an AI tool. The crucial clause is that content does not stop being AI-generated simply because a human subsequently made substantial edits. By contrast, AI-assisted includes human-created material that a tool only edited, refined, checked for errors, or helped brainstorm, where the author ultimately created the actual text or image. KDP says that AI-assisted work does not require disclosure under this rule. Its policy may change, so inspect the live help page and current publishing interface when you submit.
This is a content-origin distinction, not a judgment about writing quality. Neither “I changed most of the sentences” nor “I used a very advanced prompt” establishes that the underlying content was merely assisted. Conversely, using a tool to correct punctuation in prose you authored is not the same as asking it to generate a complete section.
Six examples: what to record before uploading
The following cases are fictional teaching scenarios, not determinations made by Amazon about actual books. Classifications follow the descriptions in KDP's published policy; borderline cases need a careful review of the actual creation history.
| What happened? | Working classification | What to retain |
|---|---|---|
| A model drafted chapter 2, then an editor rewrote its paragraphs. | AI-generated text | Draft versions and editor notes |
| An author wrote chapter 3 and used AI only for proofreading. | AI-assisted; no AI-generated text in this example | Original author draft and changes |
| An image model created the published cover illustration. | AI-generated image | Asset origin and permitted-use review |
| A human drew the cover; AI checked the color contrast. | AI-assisted; no generated image in this example | Original drawing and edit history |
| A model translated human prose into the published Spanish edition. | AI-generated translation | Source language, translated draft, human QA |
| A model proposed chapter topics; the author wrote every sentence. | AI-assisted brainstorming | Authorship notes and manuscript history |
A book can contain both AI-generated and AI-assisted material. Do not classify the whole project by the lightest-touch tool use, or assume all components share one origin. Document the specific generated content so you can answer the platform's questions truthfully. The CSV is an internal editorial record, not an alternative disclosure channel and not proof of compliance.
A seven-step pre-publication audit
- Freeze the actual edition. Give the manuscript, cover, illustrations, and translations a version identifier. If the team changes assets after the audit, revisit the changed entries before uploading.
- Inventory published material, not merely prompts.List each chapter or section, the cover, interior art, tables containing generated text, and translations. Avoid vague labels such as “AI used somewhere.”
- Trace the first substantive draft. For each asset, note who or what created its actual expressive content. Save the initial human draft where it exists and identify which output was generated. A later editor does not erase the original generation step under KDP's rule.
- Separate assistance from generation. An outline suggestion, grammar correction, or ideation pass is different from generated paragraphs, translations, or artwork. If the record is ambiguous, mark it “review needed” rather than automatically treating it as assisted.
- Review rights and factual claims independently.Disclosure does not license copied material, grant image rights, verify statistics, or clear a testimonial. Check attribution, permissions, factual evidence, and the storefront's remaining content rules.
- Enter the accurate answers in KDP. Use Amazon's current publication or republication flow. Consult the official wording at that moment. Do not substitute a disclaimer in the book description or the Scribora worksheet for the platform's disclosure.
- Retain a dated decision record. Keep the source files, human review notes, and the categories you reported with the final version. For later editions, audit newly generated or replaced text, images, and translations again.
Worked example: a mixed-origin nonfiction ebook
Imagine a fictional creator, Maya, preparing The Small Workshop Handbook. The book teaches community organizers how to plan short creative workshops. Every person, title, document, and outcome in this example is invented; this is not an account of a real Scribora project or an Amazon review.
Maya wrote her introduction from scratch, using AI to catch inconsistent terminology. She also recorded four permitted interviews and used a model to draft chapters 1 and 2 from her notes. She rewrote those chapters extensively, corrected unsupported numbers, and removed an unlicensed photograph. A designer drew the cover by hand, but a generative image tool produced a decorative illustration for chapter 4. Finally, Maya wrote an English conclusion and used an AI translation tool for the Spanish edition.
Her inventory has different outcomes. The introduction is AI-assisted editing on human text. Chapters 1 and 2 contain AI-generated text even after her editorial rewrite. The hand-drawn cover, on these facts, contains no AI-generated image. The interior illustration is generated artwork. The translation is AI-generated translation. No single checkbox called “I used AI” explains those differences as well as an asset-level log.
The decision log also includes separate release holds. Maya does not publish an interview quotation while permission is unresolved, and she checks numerical claims against the underlying attendance records. Those holds remain even when the relevant AI disclosure category is clear. She enters the appropriate generated-content information through KDP when she submits, then saves the uploaded edition and date for future reference.
A practical log might say: “Chapter 2 — base prose from generator — AI-generated text — interview permission open — hold until resolved.” The label conveys two separate facts: the work needs disclosure, and it is not ready for publication. Calling the prose “humanized” would solve neither problem.
Four traps that cause bad disclosure decisions
1. Assuming substantial editing changes the category
KDP explicitly addresses this case. A generated initial chapter remains AI-generated under its content guidelines even after substantial edits. Track the provenance instead of arguing from a percentage of words replaced. There is no published universal “edit more than X percent” exemption in the cited KDP rule.
2. Forgetting the cover, figures, or translation
An author can write every line of text and still have generated visual material. Likewise, a professionally edited AI translation does not become human-created merely because a translator reviewed it. Review each asset class separately; record the inputs and actual creation process, not only the final file extension.
3. Confusing platform disclosure with copyright ownership
KDP's publication rule and copyright registration are separate questions. The U.S. Copyright Office's January 29, 2025 report explains that copyright in AI-containing works depends on sufficient human authorship of protectable expression; prompts alone do not automatically qualify. That is a U.S. copyright analysis, not a global rule and not a substitute for KDP disclosure. A truthful KDP response does not guarantee copyright protection or permission to publish someone else's material.
4. Treating an automated export as approval
A generated PDF or EPUB demonstrates that a file exists, not that a retailer accepted it, the facts are supported, or the author owns every asset. Use the EPUB accessibility checklist for reading and navigation checks, and the AI ebook quality audit for source and claim review. These are additional gates, not replacements for the generated-content declaration.
Where Scribora fits, and where it does not
For a nonfiction book assembled from your own notes, documents, transcripts, or permitted research, Scribora can help structure an outline, prepare chapter drafts, support review, and export PDF or EPUB. Its outputs are inputs to your editorial process. Scribora is not an Amazon KDP submission tool, a legal classifier of generated content, or a guarantee of rights clearance, approval, or sales.
A useful handoff is to keep three artifacts together: the final manuscript, the source/claim audit, and this KDP decision log. If you are still planning the chapters, start with the free ebook outline template. If you have source material and want to test an editable book workflow, start free with Scribora's sample chapters. Review what the tool actually generated before making your KDP declaration.
The final pre-upload decision
Before pressing Publish, answer four questions independently: What actual text, images, or translations were AI-generated? What was only AI-assisted? Do you have the rights and evidence needed for every important element? And does the version being uploaded match the one you reviewed? An unknown answer is a reason to investigate, not a reason to silently choose “assisted.”
This KDP AI content disclosure checklistgives you an auditable process and a free worksheet, not an approval certificate. The safest efficient workflow is to identify each asset's origin early, preserve the first draft, resolve holds before publication, and answer Amazon's live questions accurately.
Primary sources checked November 17, 2026: Amazon KDP Content Guidelines (AI-generated versus AI-assisted text, images and translations; disclosures on publication or republication), and U.S. Copyright Office Part 2 announcement dated January 29, 2025. All worked examples are fictional. This guide provides editorial process information, not legal advice or an Amazon policy determination for any particular book.
