I’ve been publishing stuff for nearly 40 years and through those years I’ve sold more than 120,000 copies of my various books. Which is to say that I’m familiar with the publishing industry and I often have thoughts about major developments.
Because I’m also old enough to realize that most normal people aren’t nearly as interested as me, I’ll normally keep those thoughts to myself. But I just came across some fascinating research into how books containing significant AI-generated content are impacting the general book market. So I hope you’ll forgive me for this one exception to my rule.
The researchers - Tuhin Chakrabarty, Xinyue Liu, Jane C. Ginsburg, and Paramveer Dhillon - published a preprint of their paper: “Generative AI floods and dilutes the market for books” back in August. Tuhin Chakrabarty nicely summarized their research on X.
The study involved using the Pangram AI-detecting tool on the complete text of 14,419 self-published genre fiction ebooks available on Amazon. The researchers had access to three years of detailed private Amazon sales data from a Big Five publisher. (I had no idea such data even existed outside of Amazon itself. I’m not saying I would kill to get my hands on it. But I’m not saying I wouldn’t, either.)
The bottom line is that Amazon sales of novels with significant AI content are increasing faster than their bio-authored competitors. The titles people buy across all fiction genres are just about as likely to be written by clankers as humans. And more and more, real-world human readers seem to prefer the clankers.
We’re not talking about the countless thousands of awful AI-slop books that have been polluting Amazon book category algorithms for the past few years. Even non-experts can usually spot the inevitable spelling and punctuation errors on their landing pages before purchasing those. And I have the sense that Amazon has recently been doing a better job hiding them from sight.
Instead, the research focused on books that sold enough copies to reach the top 25 spots in their genres. These books were earning their “authors” real money - sometimes hundreds of thousands of dollars a year. And they’re pushing human-authored books out of the bestseller rankings and out of sight to potential readers.
As long as it’s possible to reliably coax an AI model into generating readable content, the business model is compelling. The idea is that a combination of clever prompts and some human reworking and reorganization of the output can produce a book that’s good enough to compete seriously on Amazon. Since the entire process can take as little as a couple of days, selling just a few hundred copies of each title can be enough to justify your time and effort.
Five years ago, as the first truly usable generative AI tools were coming on line, I did my own quick experiment. I used AI-generated content as the base for five technology-related books. If you note the publication dates on those titles, you’ll see that all five were published within two-weeks of each other.1
My effort might have been premature: the content was accurate and useful enough, but the writing was dead-boring. And I didn’t have the advantage of the professional marketing guidance available through today’s more mature AI models. In the end, all five books sold just 75 copies between them.
Today’s AI models can give you the very best state-of-the-art guidance for assembling covers, content, and marketing materials that’ll both attract customer attention and provide a satisfying read. In other words, people using AI are guaranteed access to the magic formula necessary for creating optimized, market-ready products.
Sure. If you’re looking for a seriously inspiring/depressing reading experience, then there’s currently no AI substitute for Kafka or Dostoevsky. And poetry? I can’t see a robot beating the Blake or Milton (or Kipling) experience anytime soon. But if it’s a few hours of simple entertainment with a convincing murder-mystery or engrossing contemporary fiction you’re after, then you’ll be just fine with AI.
What’s all this have to do with Canadian government policy? Well I’ve written plenty about the tens of millions of dollars of federal and provincial government subsidies aimed at companies that publish books no one reads. My guess is that there’s now even less of a chance those products will have any impact because of all the new competition from books that no one has even written.
Keep reading:
Could an Independent Publishing Industry Survive in Canada?
Over at his Substack, Kenneth Whyte has been excerpting a new book from his Sutherland House Publishing. Here’s the latest post:
Is Government Funding for Book Publishing Effective?
I’ve written about federal government arts funding in the past. Back then I highlighted the fact that you can plow billions into programs for producing art, but if no one’s interested in consuming what comes out the other end, then what do you have to show for the exercise?
Go ahead and laugh. But just remember that legacy, human-authored books - even those from mainstream publishers - seldom resemble great art.
I pulled the Department of Canadian Heritage guidelines for the Canada Book Fund to check how publisher subsidies are allocated. The federal program does not finance unread books in a vacuum. Under Publishing Support rules, a publisher must record at least $50,000 in net sales of their own titles to qualify. Base grants are calculated from 'factored sales,' multiplying verified domestic and export sales by category coefficients. An applicant's funding is simply their proportional share of total industry sales. Ottawa does not subsidize books nobody reads. It matches public dollars directly to commercial purchases.