I edit for a small indie press, and we now get three to four AI-generated submissions a week. Half of them are coherent enough to pass the first skim. The problem isn’t grammar or structure. It’s that they have no center of gravity. You read forty pages and realize you haven’t encountered a single insight that couldn’t be generated by averaging a hundred blog posts. That’s the real cost of the flood: it buries signal under statistically plausible noise.
The numbers tell the story #
Output speed: A human writer might produce 500 usable words per hour. An LLM agent can spin out 50,000 before lunch.Production cost: Formatting and cover software are free, and editing can be skipped entirely if you don’t care about the reader.Market entry: Self-publishing platforms don’t require a quality gate of any kind, so the flood hits storefronts instantly.Reader trust: When a shopper gets burned by one or two AI books, they stop browsing unknown authors — which hurts every debut novelist.
Those bullet points are fine on their own, but the deeper issue is how the incentive structure shifts. If a bad book costs $1.99 and a good book costs $14.99, the algorithm doesn't judge craft. It judges downloads and returns. The cheap flood creates a race to the bottom where “success” means optimizing for keyword coverage instead of for clarity. I've seen authors pad their metadata with "from scratch," "complete guide," and "AI workflow" tags just to stay visible — even when the content inside was written by a human.
What can we actually do about it #
I'm not anti-AI. I use Claude Code for brainstorming and structural rewrites. But there's a huge difference between using generative models as a thinking partner and using them as a full-content factory. The first preserves voice; the second erases it.
What I’d like to see is less about policing every publication and more about labeling and curation. Publishing platforms should have a disclosure toggle, and more importantly, review ecosystems should weight verified human authorship the way they weight verified purchases. That won't stop the flood, but it can build a sandbar where real writing can still find readers.
Also, readers need better instincts. If a book's title reads like a search query and the author has released forty titles in one year, that's a red flag. The market isn't going to regulate itself — the technology moves faster than any policy review cycle.
The flood will keep rising. The only question is whether we build better filters or just let the dilution become the new normal. I'd rather spend my energy on the filters.
Next Title: Self-Contained OS on a 2012 ThinkPad →