AI Is Not Simply Changing How Books Are Written. It Is Changing the Market Around Them
By Wendy Keir | EmpowerAi™
For the past few years, much of the conversation about generative AI and books has concentrated on the writing process. Can AI write a novel? Should authors use it to edit their work? Where does assistance end and authorship begin?
Those questions still matter, although a new study suggests that we also need to look beyond the individual book. Generative AI is beginning to alter the structure of the publishing market itself, particularly within self-published genre fiction.
Researchers from Stony Brook University, Columbia Law School, the University of Michigan and the MIT Initiative on the Digital Economy examined 14,419 self-published books sold through Amazon between 2023 and June 2026. They used full-text analysis to identify books containing substantial amounts of AI-generated material and matched that information with daily sales data. (arXiv)
Their conclusion is uncomfortable because it challenges one of the assumptions that has allowed the publishing industry to remain relatively relaxed about low-quality AI books. The usual response has been that readers will simply ignore them. Poor books will receive poor reviews, disappear down the rankings and eventually become commercially irrelevant.
The evidence in this study suggests that this is only partly true.
Books containing substantial AI-generated text generally achieved a smaller share of sales than books where no significant AI text was detected. Yet the AI-heavy books still reached commercial scale. Their share of sales increased over time and they began occupying more of the top-ranking positions that would previously have gone to other books. (arXiv)
This means AI-generated publishing does not necessarily need to win through quality. It can exert influence through volume.
When the number of books grows faster than the number of readers
The clearest finding is perhaps the difference between the growth in supply and the growth in revenue.
Across the period studied, the number of books recording sales within a quarter increased 19.2-fold. Quarterly revenue increased 8.9-fold. More books were entering the market and making at least some sales, although the financial market supporting them was growing at less than half that rate. (arXiv)
That does not mean publishing revenue collapsed. It means that revenue was being distributed across a much larger pool of titles. The average amount earned by each selling book fell across most of the genres examined.
For an individual author, this changes the nature of the competition. You are no longer competing only with other authors who may have spent a year or more developing a book. You may be competing with publishers capable of generating, packaging and releasing dozens of titles within the same period.
Each title can perform poorly and the portfolio can still have an effect. It occupies search space. It gives recommendation systems more material to process. It increases the difficulty of identifying worthwhile books. Occasionally, one title gains traction, and because the cost of creating the wider portfolio is so low, that may be sufficient.
The issue, then, is not merely that AI can produce mediocre books. Publishing has always contained mediocre books. The difference is the speed and scale at which they can now be produced.
Discovery is becoming a larger part of the author’s problem
Authors are often encouraged to think of the book as the product. They write it, publish it and then concentrate on persuading people to buy it.
That model becomes more difficult when the marketplace is flooded with an increasing number of titles and the available revenue is divided among them. Even a genuinely useful book can disappear quickly when algorithms have thousands of alternatives to display.
This is particularly important for expert authors. Their book may represent years of professional experience, a developed methodology and a body of thinking that has been tested with real clients. In a conventional online bookshop, however, it can appear beside a book produced in a few hours from a sequence of prompts. The marketplace does not automatically understand the difference in depth.
The author therefore needs to make that depth visible outside the book itself. Their reputation, professional work, public thinking, case studies, community and wider intellectual-property ecosystem become part of how readers judge whether the book deserves their attention.
The book still matters, although it may no longer be able to carry the entire commercial and reputational burden on its own.
Disclosure remains difficult for readers to see
Amazon requires Kindle Direct Publishing users to disclose AI-generated text, images and translations when publishing or republishing a book. It distinguishes this from AI-assisted content, where the author has created the work and used AI to refine, edit or improve it. AI-assisted content does not currently require disclosure. (Amazon Kindle Direct Publishing)
This is a sensible distinction in principle. Using AI to check grammar or question the structure of a chapter is different from asking it to create the chapter.
The difficulty is that platform disclosure does not necessarily give readers clear information at the point of purchase. The researchers reported that none of the books in their dataset disclosed the presence of the substantial AI-generated material they detected. (arXiv)
There are also limitations within the research itself. AI-detection systems are imperfect, and this is a working paper rather than a final peer-reviewed publication. It would be unwise to treat every classification as certain. Even so, the size of the dataset and the market patterns identified deserve attention. The study is less useful as a tool for accusing individual authors and more useful as evidence of a broader change in publishing economics.
The valuable use of AI may come after the book
There is another way of looking at this development.
If generative AI makes it easier to produce an unlimited number of books, then simply producing another book becomes less distinctive. The value shifts towards the quality of the original thinking, the author’s credibility and what the reader can do with the material.
This is where I believe the conversation around AI and publishing needs to become more mature. We have spent a great deal of time asking how AI can help authors create books faster. We have spent less time considering how it might help readers engage more deeply with books that already contain valuable intellectual property.
An AI Book Companion™ begins with the author’s completed thinking. The purpose is not to generate a substitute book or quietly manufacture new claims in the author’s name. The companion is built within the defined framework, language, boundaries and ethical position of the original work. It allows readers to question the ideas, reflect on their own situation and work with the author’s methodology in a more active way.
This creates a different form of value. The underlying human authorship remains central, while AI helps the reader navigate and apply it.
That distinction matters more as the publishing market becomes crowded with material generated cheaply and at scale. A book connected to a credible author, a defined methodology and a governed interactive experience is easier to recognise as part of a serious body of work. It has a life beyond its position in an Amazon ranking.
Authors may need to build for depth rather than volume
There will be pressure on authors to respond to an increasingly crowded market by producing more. More books, more posts, more newsletters and more content designed to keep them visible.
For some authors, a higher publishing frequency may be appropriate. For many expert authors, however, adding to the volume will not resolve the underlying problem. Their advantage lies in depth, specificity, experience and the practical usefulness of what they have created.
A strong book can become the foundation of a wider knowledge asset. It can support teaching, client work, licensing, speaking, community discussions and a carefully governed AI companion. Those extensions make the author’s thinking easier to use without diluting its ownership or integrity.
The emerging market appears to be separating into two broad forms of value. One is built around speed, scale and almost limitless supply. The other is built around trust, identifiable expertise and the quality of the relationship between the author and the reader.
Authors cannot control how much AI-generated material enters the market. They can make the provenance, substance and continuing usefulness of their own work much clearer.
Three Key Insights
1. Low-quality AI books can still influence the market.
They do not need to outsell the best human-authored books individually. Large volumes can absorb attention, ranking positions and a growing share of available revenue.
2. Discoverability is becoming harder for serious authors.
As the supply of books grows faster than the market, authors need stronger connections between their books, expertise, audience and wider body of intellectual property.
3. The most valuable application of AI may sit beyond book production.
Using AI to help readers explore and apply authentic expertise offers a more defensible opportunity than simply accelerating the production of additional titles.
Three Questions I’m Thinking About
1. How will readers distinguish between deeply developed expertise and convincing material produced quickly by AI?
2. Should publishing platforms make AI-generation disclosures visible to readers rather than retaining them only as internal information?
3. As books become easier to generate, will the real premium move towards authors who can offer provenance, interaction and a credible body of work around the book?
Sources
The principal research was published on 22 July 2026 as a working paper titled Generative AI Floods and Dilutes the Market for Books. (arXiv)
Amazon’s current Kindle Direct Publishing guidelines require disclosure of AI-generated content while distinguishing it from AI-assisted work. (Amazon Kindle Direct Publishing)
The United States Copyright Office continues to examine copyrightability, AI training and the market implications of generative AI through its multi-part Copyright and Artificial Intelligence initiative. (U.S. Copyright Office)
EmpowerAi™ | Real AI. Real Impact. Smarter Business Decisions.
AI Book Companion™ helps expert authors transform a completed book into a governed, interactive knowledge asset built around their own ideas, methodology and boundaries.