On the day Henry Kissinger passed away, his family was still preparing his eulogy. Dozens of biographies about him had already popped up on Amazon, some with his name misspelled on the cover, and the pages filled with endlessly repetitive nonsense. From obituary to bookshelf, in less than three hours.
This is not an isolated incident. Three years since ChatGPT's debut, AI-generated books have flooded publishing platforms like Amazon on a massive scale. By the end of 2025, the monthly number of new book listings reached 300,000.
Some people batch-produce biographies to cash in on the deaths of celebrities, some sell fake books under the names of experts, and some include potentially fatal identification methods in mushroom foraging guides. The New York Mycological Society warned the public: this is a matter of life and death.
In 2026, the defenses of traditional publishing houses were also breached. Hachette, one of the "Big Five" publishers, urgently pulled a bestselling horror novel after AI detection showed 78% of its content was suspected to be machine-generated. The French Booksellers Association forced Amazon to withdraw from the Paris Book Fair. At the London Book Fair, ten thousand writers published an 88-page blank book.
Publishing once meant someone took responsibility for the words. That promise is now failing. No chapters, no paragraphs, just the authors' signatures on every page.
In an age where AI can mass-produce text, blankness has become the weapon of the human author.
Three-Hour Biographies
Less than three hours after news of Henry Kissinger's death broke, dozens of new biographies about him appeared on Amazon. The covers looked cheap, some even misspelled his name, and the descriptions uniformly claimed "in-depth revelations" and "exclusive perspectives." Pay to open them, and the pages were filled with endless, circular drivel.
The same thing happened when Matthew Perry died. The mechanism behind it isn't complicated: black-market scripts monitor news sites 24/7, automatically triggering a process upon catching keywords like "obituary" or "death." Scrape a Wikipedia biography, feed it to a large language model to expand it into a hundred-page book, auto-generate a cover, auto-list it. The whole process might only require a human to click "confirm."
Before the family finished writing the eulogy, these AI biographies had already sold hundreds of copies.
On the day former New York Times executive editor Howell Raines died, six biographies about him appeared on Amazon. GPTZero judged one of them to have a 97% probability of being AI-generated. His family said, "They're just profiting from your grief."
The people behind these books don't hide. A "writer" named Melton has been churning out batches of biographies about recently deceased celebrities for months. Edward Tian, founder of GPTZero, judged, "Statistically, these books are almost certainly not written by humans." This is not an isolated event; it's an assembly line.
The deceased cannot protest. What about the living?
In early 2024, while promoting her memoir, American tech journalist Kara Swisher searched for new books on Amazon. At the top wasn't her book, but a fake titled Swisher: The Bulldog of Silicon Valley. 77 pages, a cover with a cheap, waxy feel, authored by "Jane Coelho"—a name that doesn't exist. Inside was not just nonsense haphazardly pieced together from Wikipedia, but also AI hallucinations fabricating many things she never did.
Swisher took out her phone and emailed Amazon CEO Andy Jassy. It started with a curse word and ended with a command: take this crap down.
Swisher is a nationally known tech journalist who can email the Amazon CEO directly. The vast majority of authors don't even have that option.
Publishing industry expert Jane Friedman discovered five fake books on Amazon using her name as the author. She complained to Amazon, which asked for the trademark registration number related to her claim. She hadn't trademarked her own name, so Amazon closed the case. It wasn't until her blog post went viral that the fake books were taken down.
Friedman asked, "What about less well-known authors who encounter this?" Another author said she had to report 29 impersonating books just last week.
The same thing is happening in the medical field. Top cardiologist Dr. Eric Topol discovered there were once 12 cookbooks and health guides being sold under his name and portrait, which he knew nothing about.
He called it "out-and-out fraud."
Amazon's complaint process is designed for trademark holders, not for authors. It assumes a name isn't worth protecting unless you turn yourself into a brand.
Journalist Sarah Loven spent three years writing her new book, interviewing over 100 people. A week after its release, at least five AI knockoffs appeared on Amazon. Loven said, "Undoubtedly, I feel violated. You want to make money after 20 minutes of entering prompts?"
Not all fakes escape detection. Independent romance author M.K. Crown's novel plainly printed a line in English: "I thought about it for 13 seconds. Here's an advanced version of your text, making Elena more likable, adding more humor, and providing a short, sexy description."
Another author, K. A. McDonald, directly used prompts asking AI to mimic a peer's style. Readers organized a review-bombing campaign on Reddit, dragging the rating down to 1.71 stars.
Impersonating biographies and knockoff novels deceive for money and trust, but some AI mushroom foraging guides publicly sold on Amazon could be deceiving for life. These books suggest readers use smell and taste to identify mushrooms. Experts have only one comment: you absolutely should not do that.
AI detection gave these books a score of 100%. The New York Mycological Society posted on social media warning the public: "Please only buy books by known authors and foragers, as it could be a matter of life and death."
The 300,000-Book Assembly Line
And all this is replicating at a speed far beyond imagination.
Brett Schickler sells perfume. At the end of 2022, he bet his wife he could write a book using AI. He opened ChatGPT, entered the prompt "write a bedtime story about a pink dolphin," used AI image software for illustrations, and four hours later, a 27-page children's bedtime story collection was listed on Amazon, soon receiving five-star reviews.
Schickler had nothing to do with writing, but that didn't stop him from becoming an author. According to data from venture capital firm a16z, after ChatGPT's launch, monthly eBook releases on Amazon tripled. By the end of 2025, new book releases skyrocketed to 300,000 per month. A four-hour joke by an outsider, multiplied by the scale of the platform, becomes a catastrophe.
The pollution has seeped into specific genres. In June 2023, independent author statistics found that about 80% of the top 100 eBooks in Amazon's "Teen & Young Adult Romance" bestseller list were suspected to be AI-generated.
Research by content verification companies shows that in 2025, 77% of books in Amazon's "Success" category were likely written by AI.
Who's making these books?
A 27-year-old named Giacomo Pietro Zuliani boasted on social networks about making $3 million by batch-generating 1,500 AI eBooks. He bluntly stated, "It makes no sense to publish a book no one wants to read." But Reddit users quickly exposed that you couldn't find books under his name on Amazon. His real profit method was selling courses, teaching others how to get rich with AI. The books were a front; the courses were the business. But the trash books churned out onto the platform by following his tutorials were real.
After the England women's football team won the European Cup, a slew of pseudo-biographies of players instantly appeared on Amazon. The covers were so shoddy they depicted a soccer ball as an American football, the books less than 50 pages, priced at £11. Former England women's captain Steph Houghton discovered her hard-earned 300+ page autobiography had been crudely imitated into a 50-page knockoff.
This trash gets through review because the review system isn't designed to stop it. Amazon KDP's automated review is good at catching two things: prohibited keywords and plagiarism.
But AI can avoid sensitive words via prompts, and excels at chewing up and spitting out fed data, generating endless, correct-sounding nonsense that traditional plagiarism algorithms can't handle.
The deeper problem lies in the incentive structure. On Kindle Unlimited, authors are paid based on pages read by customers. So AI books are often made extremely long, thousands of pages of filler, using enticing links in the table of contents to make readers jump to the end, or simply using bot accounts to inflate page reads.
Black-market practitioners form "mutual review-bombing alliances." A more advanced play is AI writing the book, AI reading it, AI writing the reviews. In this closed loop, only one thing is real: the human readers tricked into it, and the time and money they pay. The review system isn't losing to malicious intent; it's losing to its own incentive loopholes.
Amazon isn't unaware of the problem, but its response is always half a step behind. Plagiarism and trash book issues in the Kindle store date back 15 years, long before generative AI.
There was a pornography plagiarism scandal in 2012. In 2016, a journalist successfully copied and listed an entire old book, which became a bestseller.
It wasn't until September 2023 that Amazon limited uploads to 3 books per day per account. In December 2023, it required authors to disclose AI use. But 3 books a day still means an author can publish over 1,000 books a year; multiple accounts can bypass limits. The AI disclosure requirement lacks enforcement through detection, essentially giving tacit consent.
Some have turned this business into a legitimate army. Israeli AI publishing startup Spines recently raised $16 million in funding. It charges authors $1,200 to $5,000 to use AI for the entire process—editing, proofreading, typesetting, design, and distribution—promising to publish a book in three weeks. Writer Uchenna Okeoma calls it "speculators and extractive capitalists."
Science fiction writer Cory Doctorow's judgment is more direct: "The problem with these crapbooks isn't just that they divert revenue from actual authors. The problem is that they're defrauding buyers. These crapbooks exist to suck money from the inattentive, and to screw readers and authors alike."
Real authors are already voting with their feet. Writer Mark Wayland publicly stated he had to leave the Kindle Unlimited platform, seeking other, more transparent overseas channels, because his traffic and royalties were diluted to unsustainable levels by the flood of thousands of AI trash books. While Amazon requires checking a box for AI use upon upload, this disclosure was once deliberately hidden from consumers.
When a platform drives away the real and leaves behind the fake, it ceases to be a marketplace.
The most obvious choice for those "more transparent channels" Wayland turned to is traditional publishing. The Big Five publishers have editors, reviewers, and brand prestige—they should be the hardest fortress for AI-generated dreck to breach.
The Blank Book
But the fortress did not hold.
In early 2026, a horror novel titled The Shy Girl went viral on BookTok. Author Amy Ballard initially self-published; after the novel gained massive popularity on short-video platforms, Hachette, one of the global Big Five publishers, acquired it for formal publication in the US market.
It looked like an indie author's success story. Until readers on Reddit started analyzing sentence by sentence, pointing out the text's recurring hallmarks of ChatGPT. YouTuber Frankie spent nearly three hours dissecting the book segment by segment in a video, garnering over 1.2 million views. An AI detection company analyzed the entire book, concluding 78% of the content was suspected to be machine-generated.
The problem wasn't discovered by Hachette's editorial team, but by readers on Reddit and bloggers on YouTube. After The New York Times presented these findings to Hachette, the publisher announced it was withdrawing the book, canceling the US edition, and destroying the roughly 1,800 copies already sold in the UK. Ballard denied using AI, explaining: the editor she hired used AI without her knowledge.
This wasn't the only case. In January 2025, a philosophy book titled Hypnotic Politics swept through European intellectual circles. Its author, Jianwei Xun, was described as a philosopher born in Hong Kong and living in Berlin.
The book entered the top ten nonfiction bestseller lists in Italy and Germany, was reprinted three times in under 60 days, and was reportedly praised even by President Macron.
But Jianwei Xun didn't exist. The real author was Italian essayist Tommaso Debenedetti, who co-wrote the text using ChatGPT and Claude, then fabricated a complete, false biography. Low-end trash books trick algorithms; high-end trash books trick human judgment.
The first to feel the collapse within the industry were editors. A woman using the pseudonym Dora, who worked as a freelance literary editor for 12 years, eventually left the profession due to the flood of AI-generated content. "I find it a bit soul-crushing," she said, "because I got into publishing because I love books, I love art, I love humans."
She borrowed The Shy Girl from the library and felt that familiar "AI discomfort" from the first sentence. Literary agent Lauren Bieker's judgment was briefer: "Quality control has broken down."
Breakdown breeds backlash. Over 70 prominent writers jointly petitioned the US Big Five publishers to stop publishing machine-written books. Novelist Dennis Lehane was among them.
The French Booksellers Association forced Amazon to withdraw from the 2026 Paris Book Fair. Their statement directly accused Amazon of trying to "flood the market with AI-generated fake books, promoted by fake reviews, written by fake readers, and topping fake bestseller lists." Amazon responded, expressing deep disappointment at the "unfounded and misleading allegations."
As early as 2023, the renowned sci-fi magazine Clarkesworld received over 500 AI-generated trash story submissions in a single month, forcing editor Neil Clarke to close the submission channel that had been open for over a decade.
By 2026, this contamination had spread from self-publishing to every checkpoint of traditional publishing.
The most intense protest occurred at the 2026 London Book Fair. About ten thousand writers collectively "created" a book titled Don't Steal This Book. The red words were prominently printed over a black eye mask.
Open this book, and all 88 pages are filled with the names of the petitioning writers. No chapters, no paragraphs, not a single sentence. Nobel laureate Kazuo Ishiguro signed.
The back cover bore just one sentence: "The UK government must not legalize book theft to benefit AI companies."
A survey by The Society of Authors in the UK provided the footnote for this blank book: over 80% of authors' incomes have been impacted by AI; 43% of translators have seen significant income declines. The Society's chief said if AI companies continue to steal writers' works, leaving writers unable to survive, "the pages will truly become blank."
Ten thousand writers jointly published a book without words. In an era where AI can mass-produce text, the weapon human authors can wield in protest is blankness.
Some are trying to rebuild trust in more practical ways. In January 2025, the Authors Guild launched a "Human Writer Certified" program. By early 2026, over 3,000 authors had certified more than 5,000 books, expanding to non-members and publishers in March of that year.
The very existence of such a certification system is an absurd signal: the burden of proof has reversed. It's no longer AI needing to prove its eligibility for publication; it's human authors who must prove they are not machines. Writer Michael Hughes said, "It's really demoralizing that an act that brings so much joy to so many people for so long is being introduced to suspicion in this process."
An anonymous publishing industry editor left a warning: "My biggest fear is that the publishing industry is blind to how much public goodwill it has, and a few potent scandals could completely overturn it."
A deeper threat than scandal is already occurring. AI is powerful because it once ingested decades of high-quality human text. But as AI-generated dreck proliferates, the next generation of models will have to train on data described as "cyber slop."
AI generates trash books, trash books pollute training data, polluted data trains worse AI, worse AI generates more trash books. The "model collapse" long feared in academia is becoming reality. An ouroboros is eating its own tail, except what it initially swallowed was the classics of human civilization, and what it's now chewing is the digital waste excreted by itself and its kind.
The next time a celebrity dies, the biographies that appear within three hours will be even more numerous, only now their training data will be mixed with the residue of the last batch of trash.
This article is from the WeChat public account "Modern AI," author: Tutu.
