Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarGuardian→ original

One in Five Self-Published Books Has 25% or More AI-Written Text, and Publishers Have No Rules

A Stony Brook University study (not yet peer-reviewed) estimates that roughly one in five recent self-published books contains at least 25% AI-generated text. Romance author Coral Hart has published 370 novels under 21 pen names with AI assistance and says she earns "high six figures." Meanwhile, publishers are cancelling deals over suspected AI use without publishing clear rules, so authors should keep their drafts as proof.

AI-processed from Guardian; edited by Hamidun News
One in Five Self-Published Books Has 25% or More AI-Written Text, and Publishers Have No Rules
Source: Guardian. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceGuardian ↗
Editor-in-Chief
Key metric #1A class action by authors over roughly 500,000 books ended with Anthropic agreeing to a $1.5 billion settlement
Key metric #2The Pangram detector is estimated by research to be at least 95% accurate
Key metric #3In the US, text written by a machine is considered public domain

Generative AI has reached the book industry, and neither authors nor publishers know what the rules are anymore. A study by Tuhin Chakrabarty of Stony Brook University, which is still awaiting peer review, puts the scale this way: roughly one in five recently self-published books, about 20%, contains at least 25% text generated by a neural network. In other words, in those books at least every fourth section was written by a machine. Meanwhile, major publishers are cancelling deals over suspicions of AI use, yet they have not published clear policies. A literary agent quoted by the Guardian calls it the "Wild West."

The Novel Assembly Line

Dark romance author Coral Hart started out with Mills & Boon and has been self-publishing through Kindle Unlimited since 2015. The platform's algorithm rewards authors who release a new book at least once a month. According to Hart, some authors hire more than a dozen ghostwriters to keep up. She herself can write up to 10,000 words a day without AI, but with a neural network her pace has grown many times over: since 2025 she has published 370 novels under 21 pen names and, she says, earns "high six figures."

Kindle Unlimited pays for pages actually read, so readers abandon a weak book and it earns nothing. Hart is convinced that a bad storyteller cannot get a good book out of AI; you have to understand structure and reader tastes. The neural network, she says, removes friction and saves time. She now teaches courses on writing books with AI. She advises colleagues not to disclose their AI use unless the law requires it, because authors who do are harassed. After an interview with the New York Times, Hart herself was unmasked and received threats.

Cancelled Deals and Unreliable Checks

In traditional publishing, the reaction has been sharp. In March, Hachette pulled the novel Shy Girl from publication after signs emerged that author Mia Ballard had used AI. She said the text may have been inserted by an editor she had hired. The winner of the Commonwealth Prize for a short story was accused of using AI after a check with the Pangram detector; he denies it. The agents of debut author Jerry Faladé turned down a $2 million deal, and Telson Aurelien's novel was removed from the Goncourt Prize longlist. Nearly all of these stories began with readers or journalists running the text through a detector.

The problem is that a person accused of AI use can almost never prove their innocence. Academic estimates put Pangram at at least 95% accurate, with almost no false positives. But human intuition is wrong more often: suspicion tends to fall on authors for whom English is a second language and on neurodivergent writers.

Faladé points to this and hints at racism, noting that authors of colour have come under suspicion. Agents now advise clients not to send manuscripts as PDFs and to keep several versions of a draft rather than overwriting a single file. One agent described a longtime client who hates AI and tried so hard to write "not like a neural network" that her manuscript ended up full of strange words.

Law and the Absence of Rules

Publishers are also silent on the legal question. Under US law, a book must be "created by a human," and text written by a machine is considered public domain. A publisher does not get exclusive rights to such a book. Protection remains for works made with AI assistance, but machine-generated text risks being contested. Agent Natalie Lakosil doubts that the "human authorship" requirement will change in any meaningful way.

Meanwhile, authors are winning money from model developers: a class action over roughly 500,000 books ended with Anthropic agreeing to a $1.5 billion settlement. Other similar cases are ongoing. A group of 70 well-known writers asked the five largest publishers to promise that they would "never publish books created by a machine." No such promise followed. Expert Ted McIlvain argues that publishers need the courage to publish detailed policies. Agent Tess Callero suspects that publishers are waiting and leaving themselves room to profit from such books.

What It Means

For those building products around text, the takeaway is practical. A detector cannot be treated as the only evidence, because the cost of error for a person is too high. Hannah Bowman, who sits on the Association of American Literary Agents' AI committee, calls for a fair process in which the author has a chance to explain. Platforms and publishers need public rules on AI use, and tools for authors need revision history and a draft log so that they can document how a text was created.

Story

Key Questions Answered

What are the key details on The Novel Assembly Line?

Dark romance author Coral Hart started out with Mills & Boon and has been self-publishing through Kindle Unlimited since 2015. The platform's algorithm rewards authors who release a new book at least once a month. According to Hart, some authors hire more than a dozen ghostwriters to keep up.

What are the key details on Cancelled Deals and Unreliable Checks?

In traditional publishing, the reaction has been sharp. In March, Hachette pulled the novel Shy Girl from publication after signs emerged that author Mia Ballard had used AI. She said the text may have been inserted by an editor she had hired. The winner of the Commonwealth Prize for a short story was accused of using AI after a check with the Pangram detector; he denies it.

What are the key details on Law and the Absence of Rules?

Publishers are also silent on the legal question. Under US law, a book must be "created by a human," and text written by a machine is considered public domain. A publisher does not get exclusive rights to such a book. Protection remains for works made with AI assistance, but machine-generated text risks being contested. Agent Natalie Lakosil doubts that the "human authorship" requirement will change in any meaningful way.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…