Generative AI floods and dilutes the market for books by letting a handful of publishers release far more titles than readers can absorb, so the catalog grows much faster than sales or revenue ever do. That's not a guess — a July 2026 study of over 14,000 self-published titles put real numbers behind it, and the gap is bigger than most people assume.
Short answer: A 2026 study of 14,419 self-published genre-fiction titles found releases grew 38-fold since 2023 while unit sales grew only 7-fold and revenue just 9-fold. Books with heavy AI-detected text now take close to 40% of Amazon's Top-25 rank slots in some genres — not because they outsell everything else, but because there are vastly more of them competing for the same shelf space.
I test AI writing and detection tools for this site most weeks, and this is one of the first studies I've seen that measures the book-flooding problem with sales data instead of anecdotes. Researchers Tuhin Chakrabarty, Xinyue Liu, Jane Ginsburg, and Paramveer Dhillon ran full-text AI detection across 14,419 self-published genre-fiction titles sold on Amazon between 2023 and 2026, then matched the results against daily sales records through June 2026. None of the AI-flagged books disclosed it on their listings.
What's actually driving the flood
Three things line up to make this possible. First, Kindle Direct Publishing has no cap on how many books one account can upload, and a generative model can draft a full novel-length manuscript in hours instead of months. Second, Kindle Unlimited pays authors per page read rather than per sale, which rewards volume over any single book's quality. Third, Amazon’s AI-content disclosure rule only requires authors to flag AI-generated content to Amazon during upload — it isn't shown to shoppers on the book's product page. That's part of why the study above found zero visible disclosure across its entire 14,419-title sample: the honesty mechanism exists, but it's invisible to the reader deciding whether to buy.
The numbers behind the flood
The study's title-versus-sales split is the clearest evidence that volume, not popularity, is doing the work. Books with substantial AI-detected text (over 25% of the manuscript) make up a fifth of titles but pull a noticeably smaller share of what actually sells.
| Segment | Share of titles | Share of unit sales | Share of revenue |
|---|---|---|---|
| No detected AI text | 62.9% | 71.7% | 72.5% |
| Substantial AI text (>25%) | 20.0% | 12.1% | 11.3% |
The growth math tells the same story from a different angle. Between January 2023 and March 2026, the number of released books grew 38.3-fold, books with any observed sales grew 19.2-fold, unit sales grew 7.3-fold, and revenue grew 8.9-fold. Revenue per selling book fell in six of eight genres the researchers tracked, concentrated in genres where AI adoption was highest. The one bright spot: in Fantasy/Supernatural/Horror, where AI arrived latest, revenue per book for human-written titles actually rose 35% over the same period.
How to respond to the AI book flood
Whether you're buying books or publishing them, the fix isn't waiting for Amazon to solve this — it's changing a few habits.
1. Treat a missing disclosure as information, not an oversight
Since KDP's AI disclosure isn't shown on listings, a book's product page will never say "AI-generated" even when the policy required the author to flag it internally. Assume silence tells you nothing either way, and look for other signals instead — author history, review patterns, and sample-read quality.
2. Read the sample before you buy, not just the blurb
Blurbs and covers are the easiest part of a book to make polished; the actual prose is where AI volume shows up as repetitive sentence rhythm and vague, interchangeable detail. Amazon's "Look Inside" or a free sample chapter takes two minutes and catches more than star ratings do.
3. Run a suspicious excerpt through a detector as a second opinion
When I ran a handful of top-ranked self-published openings through GPTZero and Originality.ai this week as part of testing detection tools, both flagged clear AI patterns in listings carrying no visible AI notice at all. A detector score isn't proof — our guide to spotting AI writing covers the false-positive risk — but it's a useful second check once your own read already feels off.
4. If you're a self-published author, label your AI use instead of hiding it
The Authors Guild’s Human Authored certification exists specifically so human-written work can stand out from a catalog that's growing 38 times faster than four years ago. Readers increasingly search for that signal, and undisclosed AI-generated content that violates KDP's terms risks a blocked listing or account suspension.
5. Watch page-reads and return rates, not just star counts
In Kindle Unlimited-heavy genres, books with no AI text lost 8.5 percentage points of sales share and 8.4 points of revenue share compared to lower-KU genres, per the study. If you're an author competing there, page-read completion and refund rates are now a better early signal than ratings, which lag and are easy to pad.
Prompts you can use to check a book or your own manuscript
These work with any general-purpose AI assistant. Paste a sample chapter or your own draft in place of the brackets.
- Check for AI-typical patterns: "Read this book excerpt and list any phrases, sentence-length patterns, or structural habits that are typical of AI-generated fiction: [paste excerpt]."
- Check for lived, checkable detail: "Which sentences in this excerpt contain a specific detail a human author would need firsthand knowledge to write, and which are generically plausible for any book in this genre? [paste excerpt]"
- Draft an honest disclosure statement: "I used an AI assistant to draft roughly [X]% of this manuscript and edited the rest by hand. Write a short, honest author's note disclosing this to readers."
- Self-check before you publish: "Rewrite any paragraph in this manuscript that reads as templated or AI-generated, and tell me exactly which ones you changed and why: [paste draft]."
Common mistakes to avoid
The biggest mistake readers make is trusting star ratings alone — a book can rack up early five-star reviews from a small pool of buyers before enough people finish it to notice the writing is thin. The biggest mistake authors make is assuming a detector-free upload means the disclosure obligation goes away; KDP's rule applies whether or not any tool would flag the text, and enforcement has tightened through 2025 and 2026. A third mistake is treating "AI-assisted" and "AI-generated" as the same thing — KDP’s own policy only requires disclosure for content an AI tool actually produced, not for brainstorming, grammar checks, or editing your own draft. A fourth is publishing volume as a strategy without editing; the study found rare-expression overlap with existing books actually rose alongside revenue for AI-heavy titles, which reads less like originality and more like recycled phrasing that happened to sell anyway.
Tools that make this easier
Detection and disclosure are two separate problems, and different tools help with each. For checking whether a sample reads as AI-generated, our guide to spotting AI writing walks through the same detector workflow used above. If you're an author who wants AI help with drafting or brainstorming without crossing into fully AI-generated content, see our picks in best AI writing tools and best AI tool for content creation — both cover which tools keep you as the primary author versus which ones write the whole draft for you. For a broader look at how we test any AI tool's real-world claims before recommending it, our AI tool reviews hub is the place to start. And if the disclosure debate reminds you of the EU's parallel push on labeling, our coverage of the EU’s mandatory AI-content labeling rules shows where that regulatory pressure might head next for retailers like Amazon.
My take
I don't think the answer is banning AI-assisted books — plenty of working authors use AI for brainstorming or line edits without producing anything close to the volume this study is describing. What actually needs to change is visibility: KDP already collects an AI-generated disclosure at upload, and there's no good reason it can't show up on the product page the way "translated by AI" or edition details already do. Until that happens, the honest move for readers is to stop treating star ratings as a quality signal in fast-moving genres, and the honest move for authors is to opt into something like the Human Authored certification rather than wait for Amazon to force the issue.
Frequently Asked Questions
Is it illegal to sell AI-generated books on Amazon?
No. Amazon allows AI-generated books, but its KDP policy requires authors to disclose AI-generated text, images, or translations when they publish or edit a listing. Selling undisclosed AI-generated content violates KDP's terms and can lead to a blocked book or suspended account, but it isn't a legal violation on its own.
How can I tell if a book I'm considering is AI-generated?
Read the free sample before you buy — AI-heavy text tends to have repetitive sentence rhythm and generic detail that a human author's firsthand knowledge would naturally avoid. You can also run a short excerpt through a free detector like GPTZero as a second opinion, though no detector is proof by itself.
Does Amazon show AI disclosures to shoppers?
No. The disclosure authors give Amazon during the KDP publishing flow is collected internally and isn't displayed on the book's product page, which is one reason the 14,419-title study behind this article found no visible AI disclosure across its entire sample.
Is this problem getting better or worse?
Worse by the numbers so far: AI-detected books' share of Top-25 rank slots rose from near zero in early 2023 to roughly 40% by the second quarter of 2026. Whether it improves depends largely on whether Amazon makes disclosure visible to readers and whether more authors adopt voluntary labels like the Authors Guild's Human Authored certification.