Publishing Doesn't Fear AI. It Fears Volume.

In September 2023, Amazon made a call it had never had to make in twenty years of running an online bookstore: capping the number of books an author can publish per day. The limit was set at three. A ridiculous ceiling for a human, who doesn't write three books before lunch. The move wasn't aimed at humans.

The trigger at the time had been an AI-generated book about the Maui wildfires, put up for sale a few days after the disaster, which "smelled like AI" according to the first readers. Three years later, the phenomenon has gone industrial.

That's the whole problem generative AI now poses to the publishing world. An op-ed relayed by Le Monde Pixels on June 21 shifts the debate to where it should have been all along. The question is no longer whether a machine can imitate an author's style or write a decent novel. The question is volume.

The wrong case: quality

For two years, the debate about AI and creation revolved around a single fear: what if the machine wrote as well as we do? People compared paragraphs, hunted for ChatGPT's tics, sneered at flat sentences and recycled metaphors. Reassuring, because there was always something to point at.

Except that case misses its target. In a distribution market, the average quality of a title matters far less than its ability to exist, to show up in a search, to occupy a slot on a virtual shelf. And to occupy a slot, a mediocre book does the job perfectly. It just has to be there, in numbers.

The right case: marginal cost

Here's the mechanism. Writing a 30,000-word guide on any subject used to take a human author weeks. With a generative model, it's a matter of hours, and the second version costs even less than the first. The cost of producing one more title has fallen close to zero.

When producing costs nothing, you produce without limit. A single synthetic publisher can put out fifty titles a month, spread them across adjacent categories, buy a handful of reviews, and generate enough noise to bury a legitimate author at the bottom of the results. In 2026, more than 30% of new titles filed on Kindle Direct Publishing carry some form of AI disclosure.

It's an attention economy turning against itself. The value of careful editorial work rested partly on its scarcity: few people knew how to write a good guide, it took time, so it was worth something. Flood the catalog with free, infinite copies, and that scarcity evaporates. Human work doesn't become bad. It becomes impossible to find.

The flood has named victims

This machinery is anything but abstract. Ask Jane Friedman, author and a well-known figure in advising writers, who discovered in August 2023 half a dozen books sold under her name on Amazon. She had written none of them. AI-generated titles, pinned to her reputation to sell better. Amazon first refused to take them down, since she couldn't show a registered trademark on her own name.

Travel guides were a favorite hunting ground. A certain "Stuart Hartley" moved a series of tourist guides, each loaded with 400 to 500 glowing reviews written in the same chatbot style. Cooking, gardening, programming, medicine: no niche was spared.

And sometimes the flood literally kills. In the autumn of 2023, the New York Mycological Society warned about AI-written mushroom foraging guides sold on Amazon under invented author names. Some recommended tasting a mushroom to identify it. Most of the deadliest species look like common edible ones. A fake guide drowned among the real ones, and the result is poisoning.

Don't fight the wrong battle

Just don't mix everything up. The major open letter signed by 76 cultural organizations, carried in France by the Syndicat national de l'édition, mostly targets model training and copyright: transparency about ingested works, respect for intellectual property, a removal mechanism. It's a legitimate fight, but a distinct one. You can perfectly well settle the training question and keep being swamped by the volume produced downstream.

The letter does recall what's at stake: cultural industries account for 4.4% of the European Union's GDP and 7.5 million jobs. A sector that size doesn't collapse all at once, it dilutes.

Conversely, reassuring yourself by repeating that AI writes badly, so good books will eventually rise to the top, is denial. On a saturated channel, mediocrity wins by sheer numbers. Readers have neither the time nor the tools to sift through fifty cloned guides to find the one written by a human who actually knew the subject.

The danger to publishing therefore doesn't come from what the machine might produce that's more beautiful or more accurate. It comes from what it produces in quantities no one can absorb, in a space that was already finite. What's threatened isn't writing. It's being found.