Guide

The AI book flood, and how one good book stands out

In 2023 a science-fiction magazine stopped taking submissions because machine-written stories outnumbered the real ones. Since then every platform has built dams. Here is what they look like, and why they favour authors who write one book well.

Neubook Write · 5 September 2026

What the flood looked like

Clarkesworld editor Neil Clarke closed submissions in February 2023 after rejecting more than 500 machine-written stories in three weeks, against roughly 700 legitimate ones. He traced the wave to side-hustle videos telling viewers to paste ChatGPT output into paying markets. In the same month Reuters counted more than 200 Kindle ebooks listing ChatGPT as author or co-author, and a later analysis found that books with no detected AI text earned less as AI titles crowded into their categories.

Distributors felt it too. Draft2Digital's Dan Wood told Jane Friedman that inbound volume in 2024 ran about 50% above normal and that bad actors were easy to spot because "normal authors don't release 10 books in a day".

The dams

  • Amazon capped new titles at three per account per day in late 2023 and requires an AI content declaration on every submission. It has since reviewed back catalogues and removed undisclosed titles.
  • Draft2Digital, IngramSpark and Barnes & Noble Press watch account behaviour (upload cadence, duplicate text, cover reuse) and have tightened content guidelines.
  • Magazines and competitions now mostly refuse AI-drafted work outright. The Authors Guild's Human Authored mark exists to signal the opposite.
  • Search and store algorithms demote mass-produced sets of near-identical pages and listings; see Google's scaled content abuse policy.

Every one of those measures targets volume and anonymity. None of them targets an author who publishes one honest, edited book with their name on it.

Five things that make one book stand out

  1. A reader you can name. "Youth pastors in their first year" beats "Christians". "People who killed three sourdough starters" beats "home bakers". The flood is generic by construction; specificity is the moat.
  2. Experience the model does not have. The scene you were in, the number you measured, the mistake you made. Put one in every chapter.
  3. A voice. Give the tool a sample of your writing and edit toward it. The research on homogenisation shows this is where AI-assisted books lose ground if nobody intervenes.
  4. Finish and edit before you publish. The flood is unedited. A book that a human read twice is already in the top tier of its category.
  5. Disclose. It removes the one risk that can end an account and signals that you are not the thing the platforms are fighting.

The opportunity inside the flood

Readers have become good at spotting the generic and are actively looking for the specific. Reviews now praise books for "sounding like a real person". Small, particular audiences, such as a language community, a denomination, a trade, or a town, are underserved because the flood does not bother with them. That is where a first book by someone who actually belongs to the audience can find its readers.

One book, written with you. Neubook Write asks who the book is for before it plans a chapter, drafts from your answers and your voice, and gates nothing on the first book. Read it and download it free.

Start a book, free

Sources

Related

Do AI detectors actually work?GuideCan you copyright a book written with AI?GuideThe EU AI Act and your AI-assisted bookGuideThe words that give AI writing awayGuideDoes AI make writing better? What the studies actually foundGuideAll guidesPublishing with AI, honestly