Amazon requires you to disclose AI-generated content when publishing through KDP. Since September 2023, the upload process includes a mandatory AI disclosure section. Getting it wrong can result in your book being removed, your account flagged, or, in extreme cases, your publishing privileges suspended.
Here's what Amazon actually enforces, where the gray areas are, and how to publish AI-assisted work safely.
Amazon's AI Disclosure Policy: The Rules
Amazon distinguishes between two categories:
AI-Generated Content
Content where AI produced the text, images, or other creative elements. This includes:
- Text drafted by AI (ChatGPT, Claude, Sudowrite, etc.) even if you edited it afterward
- Images created by AI tools (Midjourney, DALL-E, Stable Diffusion)
- Audio narrated by AI text-to-speech for audiobooks
Requirement: You must disclose AI-generated content during the KDP upload process.
AI-Assisted Content
Content where AI was used as a tool, but a human created the output. This includes:
- Using AI for brainstorming or research
- AI-powered grammar and spell checking
- AI editing suggestions that you implement manually
- Using AI to generate ideas that you then write yourself
Requirement: No disclosure required for AI-assisted text. AI-assisted images still require disclosure if the image itself was AI-generated.
The Disclosure Process on KDP
During the KDP upload flow, you'll encounter the AI Content section. Here's what it asks:
| Question | Options | When to Select |
|---|---|---|
| Does your book contain AI-generated text? | Yes / No | Yes if AI drafted any significant portion |
| Does your book contain AI-generated images? | Yes / No | Yes if any cover art or interior images are AI-generated |
| Describe how AI was used | Free-text field | Brief, honest description of your process |
Important: This disclosure is visible to Amazon's review team but not to customers. There is no public-facing AI label on your book listing (as of early 2026).
What Amazon Actually Enforces
Amazon's enforcement has evolved significantly since the initial policy launch. Here's what we know from author reports, KDP communications, and observed patterns through early 2026.
Active Enforcement
Mass-produced AI content: In late 2023, Amazon began removing books that appeared to be bulk-generated AI content, often published under fake author names at a rate of 5-10 titles per day. By mid-2024, KDP implemented upload rate limits (3 books per day per account) partly in response to this flood.
Undisclosed AI images: Amazon has been more aggressive about AI image detection than text. Books with AI-generated covers that weren't disclosed have been flagged and temporarily removed from sale. Authors report receiving emails requiring them to update their disclosure before the book is republished.
Quality-based removal: Amazon has removed AI-generated books that received multiple quality complaints from readers, regardless of disclosure status. This suggests that disclosure alone doesn't protect against quality-based enforcement.
Gray Areas
Edited AI text: If AI drafted your text but you substantially rewrote it, is it "AI-generated" or "AI-assisted"? Amazon hasn't published a clear threshold. The practical consensus among publishing attorneys: if a reader could independently recreate the essential content by prompting an AI themselves, it's likely "generated." If your expertise, voice, and original insights are the primary value, it's likely "assisted."
AI-translated books: Using AI to translate a book you originally wrote in another language falls into a gray area. Most authors disclose AI translation as "AI-generated" to be safe.
AI-formatted content: Using AI to format or structure existing content (e.g., AI-generated table of contents, AI-suggested chapter breaks) is generally considered "assisted" and doesn't require disclosure.
What Hasn't Been Enforced (Yet)
- AI editing tools: No author has reported enforcement action for using Grammarly, ProWritingAid, or similar editing tools without disclosure
- AI brainstorming: Using AI to develop ideas, outlines, or character concepts hasn't triggered enforcement
- Hybrid workflows: Authors who use AI for first drafts but substantially rewrite haven't reported issues when disclosing as "AI-assisted"
Real Consequences: What Happens When It Goes Wrong
Account Warning
The most common outcome for first-time disclosure violations. Amazon sends an email requiring you to update your AI disclosure. Your book may be temporarily unpublished during review. Most authors resolve this within 48-72 hours.
Book Removal
For repeated violations or books with quality complaints. The book is removed from sale, and you receive a content quality notification. You may republish after updating disclosure and potentially addressing quality issues.
Account Suspension
Reserved for egregious cases, typically mass-publishing clearly AI-generated content without disclosure, especially when combined with misleading authorship claims. Account suspensions are rare but devastating: you lose access to all your titles, royalties in pending payment may be delayed, and reinstating a suspended account is a months-long process.
Ranking Suppression
Separate from the AI disclosure system, Amazon's content quality checks can suppress your book's visibility for technical accessibility issues in your EPUB file. This happens regardless of whether your book is AI-written or human-written, and it's the most common silent penalty authors face.
Missing alt text, broken heading structures, and absent accessibility metadata trigger Amazon's quality flags. Your book stays listed, but it effectively disappears from search results. For details on how this works and how to fix it, see our Amazon suppression guide.
The Financial Reality of AI-Generated Books
Let's address the elephant in the room: can you make money publishing AI-generated books on Amazon?
The Numbers
A 2025 analysis by K-lytics of 5,000 books with suspected high AI content found:
| Metric | AI-Generated Books | Human-Written Books (same genres) |
|---|---|---|
| Avg. monthly revenue (after 6 months) | $47 | $312 |
| Avg. reviews (after 6 months) | 3.2 | 18.7 |
| Avg. review rating | 3.1 stars | 4.2 stars |
| Return rate | 14.8% | 4.2% |
| Still in publication after 12 months | 34% | 78% |
The data is clear: AI-generated books underperform human-written books in the same genres by a wide margin. The exceptions are low-content books (journals, planners, puzzle books) and highly specialized nonfiction where the information itself is the value.
Why AI-Generated Books Underperform
- Reader detection: While individual paragraphs may pass, full book-length AI content has detectable patterns, repetitive phrasing, emotional flatness, lack of genuine insight
- No marketing engine: Successful self-publishing requires an author brand, email list, and consistent marketing. Publishers of mass AI content rarely invest in these
- Review quality: AI-generated books attract more negative reviews, which tank Amazon's recommendation algorithm
- Return rates: Higher return rates signal quality issues to Amazon's algorithms
How to Publish AI-Assisted Books Safely
The Safe Workflow
-
Use AI as a tool, not a substitute. Brainstorm, outline, draft sections, and edit with AI, but ensure your expertise, voice, and original contributions are the book's core value.
-
Disclose honestly. When in doubt, select "Yes" for AI-generated content and describe your process. Over-disclosure is never penalized; under-disclosure can be.
-
Invest in quality. A human editor, professional cover design (or properly disclosed AI covers), and thorough proofreading separate legitimate AI-assisted publishing from content spam.
-
Build an author brand. Use your real name (or a consistent pen name), engage with readers, and produce books that reflect genuine expertise or creative vision.
-
Check your EPUB's technical quality. This is the step most authors skip. Amazon evaluates your file's accessibility independent of its content quality. Run an accessibility scan before uploading.
Disclosure Language That Works
For the KDP free-text disclosure field, clear and specific language is better than vague statements:
Good examples:
- "AI tools (ChatGPT) were used for brainstorming and first-draft generation. All content was substantially rewritten, fact-checked, and edited by the author."
- "Cover image was generated using Midjourney and edited in Photoshop."
- "AI was used to assist with research summarization and outline development. All text was written by the author."
Bad examples:
- "Some AI may have been involved" (vague)
- "AI-enhanced" (meaningless)
- No description at all (when you selected "Yes")
International Context: Beyond Amazon
Amazon isn't the only platform with AI content policies, and it isn't operating in a regulatory vacuum.
Platform Policies (2026)
| Platform | AI Text Policy | AI Image Policy | Public Disclosure |
|---|---|---|---|
| Amazon KDP | Mandatory disclosure | Mandatory disclosure | No (internal only) |
| Apple Books | Quality-based review | Mandatory disclosure | No |
| Kobo | No specific policy | No specific policy | No |
| Google Play Books | Mandatory disclosure | Mandatory disclosure | Optional |
| IngramSpark | No specific policy | No specific policy | No |
Regulatory Landscape
EU AI Act (effective August 2025): Requires marking AI-generated content in commercial publications distributed in EU markets. This applies to text and images. Enforcement mechanisms are still developing, but publishers distributing in the EU should treat this as legally binding.
US: No federal AI content disclosure law as of early 2026, though several states have proposed legislation. The FTC has signaled interest in AI disclosure requirements for commercial content but hasn't issued formal rules.
UK: The UK's approach mirrors the EU's in principle but with lighter enforcement. The Online Safety Act's provisions on AI content are being interpreted to include published works, though case law is limited.
The Quality Question
The most important factor in publishing AI-assisted books isn't the disclosure, it's the quality. Amazon's algorithms, reader reviews, and return rate tracking don't care whether you used AI. They care whether readers are satisfied.
Books that succeed long-term, regardless of how they were created:
- Deliver genuine value (entertainment, information, perspective)
- Reflect a consistent, distinctive voice
- Meet technical quality standards (formatting, accessibility, error-free text)
- Come from an author who engages with their audience
Books that fail, regardless of how they were created:
- Read like generic filler
- Contain factual errors or logical inconsistencies
- Have technical issues that impair the reading experience
- Come from an anonymous, disengaged publisher
AI is a tool. Like any tool, its value depends entirely on the skill and intention of the person using it.
The Accessibility Angle Most Authors Miss
Here's a scenario we see regularly: An author spends weeks crafting an AI-assisted book. They disclose properly, hire an editor, design a professional cover. They upload to Amazon feeling confident.
Three months later, sales plateau well below expectations. The book doesn't appear in relevant search results. The author assumes it's a marketing problem.
It's not. It's an EPUB problem.
Amazon's automated quality system evaluates every uploaded EPUB for accessibility compliance. Missing alt text on images, invalid heading hierarchy, and absent accessibility metadata trigger suppression flags. Your book is still technically for sale, it's just invisible to most of your potential readers.
This has nothing to do with AI content policies. It applies equally to books written entirely by hand. But authors focused on the AI disclosure question often overlook the technical quality of their EPUB file.
The fix is straightforward: scan your EPUB for accessibility issues before uploading. Tools like Rahatt identify these problems in seconds, and most fixes take minutes. Learn more in our EPUB accessibility guide.
FAQ
Will Amazon ban AI-written books?
No indication suggests a complete ban is coming. Amazon's approach has been to require disclosure and maintain quality standards, not to prohibit AI use. What Amazon does remove is mass-produced, low-quality content that degrades the customer experience, regardless of whether AI was involved.
Do I need to disclose using Grammarly or ProWritingAid?
No. Amazon considers grammar checkers, spell checkers, and similar editing tools as standard writing aids, not AI content generation. No disclosure is required for using these tools.
Can Amazon detect if my book was written by AI?
Amazon hasn't publicly confirmed using AI detection tools on submitted manuscripts. However, they review books flagged by quality complaints, return rates, and automated content analysis. The practical risk of detection increases with the proportion of unedited AI text in your book. For a deeper look at detection capabilities, see our article on what readers can tell about AI writing.
What if I used AI for my first draft but rewrote everything?
This is the most common gray area. If you substantially rewrote the content, adding your own expertise, voice, and original material, most publishing attorneys advise disclosing as "AI-assisted" rather than "AI-generated." Use the free-text field to describe your process honestly. The key question: is the final text meaningfully different from what the AI produced?
Should I disclose AI use to my readers?
Amazon's disclosure is internal (readers don't see it). Whether to tell readers is a separate decision. The trend in 2026 leans toward transparency: a brief author's note explaining your process builds trust rather than eroding it. This is especially important for nonfiction, where credibility depends on the reader trusting your expertise and process.