Creative Markets are being Swamped

Last week Publishers Weekly reported that the number of books published with ISBN numbers in the US in 2025 had jumped by 32 percent over the previous year to more than 4 million titles.

The increase was led by self-published works, for which the number of print and e-books … soared 38.7% to more than 3.5 million from 2.5 million in 2024… The number of traditionally published books released in 2025 rose 6.6%, to 642,242, from 602,428 in 2024. 

That’s an enormous increase, even accounting for the fact that books often have multiple ISBN numbers tied to their different formats (ISBN’s are unique identifier numbers issued by Bowker, that allow published books to be tracked).

A spokesman for Bowker, Andrew Kovacs, told PW that:

“The fact that every aspect of the publishing process once available only through traditional publishers can now be obtained from the self-publishing service providers at a comparable level of quality” is likely the key driver in lifting output.

Certainly true. Several friends writing books in the past year have chosen to go the self-publishing route for a variety of reasons. There is still prestige value in publishing with a traditional publisher but traditional publishers have scaled back promoting titles and providing rigorous editing.

Given the low cost of print-on-demand and the army of publication services that can guide authors through getting their books published and into distribution, the process has streamlined. Perhaps a similar transition to the early days of blogging in the early 2000s when digital publishing platforms gave anyone with something to say a cheap easy way to reach the world. This is the democratization of culture argument — make tools easier and cheaper and more people will have access to make more. And the marketplace will have a richer pool to choose from.

But Journalist Bob Sacks wonders how much AI-written books have contributed to the flood and points out there’s currently not a way to definitively determine it, pointing to a structural problem in how we account for creative work in marketplaces. This appears to be true across creative marketplaces:

• A study in July out of Columbia Law School reported that of “14,419 self-published genre-fiction titles on Amazon 2023–2026: about 20% had substantial (25%+) AI text, 2,168 were majority AI, nearly 1,000 were 90%+ AI.”
• In June the music-streamer Deezer reported that 90,000 AI-created music tracks were being uploaded to its platform every day, accounting for more than 50 percent of total uploads.
• Robert Kneschke reported on his blog Daily Life of a Photo Producer that the number of AI images in Adobe Stock’s portfolio was almost 50 percent — 313 million AI images versus 342 million photographs as of April 2025, up from 2.5% of the library in May 2023, with 29 million AI images accepted in January–February 2025 alone.
• And Graphite reports that a Common Crawl sample in May of 55,000 web articles run through AI detectors indicate that almost 50 percent of the articles were primarily written by AI. All you have to do to confirm is look at your LinkedIn feed. Substack recently introduced a Pangram feature that allows readers to check how much AI was used in a post.
• Perhaps even more insidious are reports that books are showing up in online sellers that purport to be written by established writers but in fact are written by AI, thereby stealing fans of those writers (and likely disappointing them).

AI is distorting and swamping creative marketplaces everywhere. When AI can spin out 200 Hemingway novels before lunch or a thousand songs before dinner, the harm to markets where artists sell their work is devastating. It’s not so much that AI makes better work (yet, at least) or more cheaply (which it already does), it’s that it swamps and distorts supply so much that it makes it almost impossible for artists to be found or sell enough to be make a living. And because recommendation algorithms are optimized to engagement and AI can optimize its content flood to optimize for the algorithms, human work not similarly optimized becomes invisible.

If your work is primarily consumed through streaming media, your income is likely being diluted because of the flood. And if you’re a consumer, how do you navigate through the abundance, particularly now that human curators — academies, critics, publishers, editors — are diminished as algorithms increasingly determine what gets shown? The effort to sort and choose has exacted an escalating cognitive toll that at some point suppresses the desire to read, to listen, to watch.

This exhaustion is increasingly expressing itself in reflexive rejection of AI. When the cost of creating something plummets and a marketplace floods with abundance, monetary value usually craters too, displacing those who used to supply that market. In this case, skilled artists and those who get artistic work to market.

The instinct is to build better, more accurate ways of identifying what was made by humans because that work has value. But then what, lock out the machine-made? How about the increasing number of artists using AI tools in their work? (one survey reports that two-thirds of commercial musicians admit using AI) Is art made by artists using AI tools not art? If so, where’s the line between how much is human and how much is machine? Is art valuable because of who or what created it or because it communicates something?

Another complication is that AI-detectors are frequently unreliable and that human-authored work is often tagged as AI when it is not. Enormous efforts are underway right now to create and mandate ways to determine whether creative work is made by AI or not. In August an EU law went into effect requiring disclosure when AI is used. Anthropic responded by embedding watermarks invisible to humans into its outputs. Of course in a matter of days the internet was full of ways to break the watermarking. This is the kind of whack-a-mole game played 25 years ago when the recording industry tried to embed Digital Rights Management codes into music to stop piracy, but each new generation of protection was quickly subverted by pirates.

One might argue that the vast majority of self-published books and music and images in traditional marketplaces is already of lesser, derivative quality. That hasn’t stopped demand for it. And might AI not help those creators produce better work, raising the level? YouTube’s shaky cameras and poor lighting when it launched didn’t stop an audience. The quality of video-making got seriously better as more people got better at using the tools. And it changed the definition of what good video is. Ditto images and Photoshop.

So no easy answers. To go back to my earlier question: Is art valuable because of who or what created it or because it communicates something that moves us? I suspect the answer is less clear than most of us would admit. Artists would surely argue it’s both, but the marketplace — where do-I-like-this? is the ultimate test — may have a different answer.

Also Worth Your Attention

The audience willing to pay more — but for what?: The British Museum sold £2.4 million of Bayeux Tapestry tickets in a single day, a record sale that overwhelmed projections. Then, after the show opened the museum banned photographing inside the galleries because selfie-takers were clogging the visitor flow. Actress Cate Blanchett stepped out of the curtain call at London’s National Theatre to insist the audience stop filming. Then there’s the story of Toronto dance clubs banning cameras and smart glasses from the dance floor. Meanwhile, a study by Goldman Sachs reports that “superfans” generate an extra $4 billion a year of revenue. American movie theaters had their biggest summer ever at the box office, taking in $4.76 billion. And a UK startup has raised £15 million with the intention of recreating classic concerts with digital avatars in unique spaces, not the original ones, perhaps inventing a whole new genre of entertainment experience. So what does this say about the culture value transaction? Is it the object or performer? Or is it an experience that can be modified, remixed or synthesized? There seems to be a lucrative audience for that. But if it’s simply being there, in a theatre or in a gallery, where your full attention is demanded, our devices are competition, and increasingly, venues will treat it that way.

Who should pay for culture? The Wall Street Journal makes a case for privatizing the Smithsonian. Why? To insulate itself from government interference: a business model that doesn’t involve government money. This in a week that respected Smithsonian chief Lonnie Bunch announced his retirement and the Department of the Interior posted signs around the American History museum, accusing it of bias (!). But is private funding that much better? Josh Goldblum, writing about the Met Museum’s decision to cancel a controversial John Galliano retrospective, described the private-funding trap: “Donors underwrite celebration,” he writes. “Nobody endows a wing to say ‘it’s complicated.’” The Met’s glamorous Costume Institute gala raised a record $31 million this year. Do you think none of that money came with strings? Public money comes with politician interference while private money comes with sponsor expectations. Neither is truly independent. So who pays for “it’s complicated”?

Related