Aller au contenu principal
General

Fake AI Books: What the Authors' Open Letter Actually Says

The European Writers Council and 1,100+ U.S. authors ask publishers to ban AI-generated books. The debate is already drifting — here's the line, and why it's called "vibe writing".

12 days ago8 min read

"The writing that AI produces feels cheap because it is cheap. It feels simple because it is simple to produce. That is the whole point."
— Open letter from 70+ authors to literary publishers, June 2025

Two coordinated appeals have put generative AI on every serious publisher's desk over the past months. In May 2025, the European Writers Council (the European federation of authors, representing 220,000 professional writers across 32 countries), together with CEATL (the European Council of Literary Translators' Associations) and FEP (the Federation of European Publishers), issued a joint statement asking EU Member States and the Commission to protect human-created works, label AI-generated content, and reserve cultural public funding to books written by humans. In late June 2025, an open letter signed by more than 70 authors — Dennis Lehane, Gregory Maguire, Lauren Groff, Jodi Picoult, Paul Tremblay among them — addressed the Big Five U.S. publishers (Penguin Random House, HarperCollins, Simon & Schuster, Hachette, Macmillan) demanding they pledge never to release books "created by machines." The accompanying petition gathered over 1,100 signatures in less than 24 hours.

The numbers are clear. The position is clear. I support it.

I'll note one detail anyway. HarperCollins, which appears on the list of publishers being asked not to release AI books, signed a deal with Microsoft in November 2024 to license its nonfiction backlist for AI training at $5,000 per title. So a HarperCollins author signing the petition is asking their own publisher for something that publisher already invoices, to a named buyer. That's not a reason not to sign — if anything it's an extra reason to. It's just a reason to keep your eyes open about the internal contradictions of the camp you're joining.

Three writers, the same word, three very different practices

Let me detour through three real people.

Writer A is finishing a crime novel. She uses ChatGPT to check the spelling of a Basque surname she heard once, to confirm whether a Vienna-to-Trieste train existed in 1903, and to find a synonym when she's stuck on a word for five minutes. Everything else — the sentence, the rhythm, the breath between two lines of dialogue, the decision to cut a chapter — is hers. She writes, let's say, 100 % of her text.

Writer B is working on a fantasy novel. He uses Sudowrite to generate a chapter from a scene outline. He keeps 70 % of what comes out, rewrites the rest by hand, adds his humor, removes what rings hollow. He writes, let's say, 30 % of his text. The other 70 % is a co-production.

Writer C doesn't bother. He writes a three-paragraph prompt, runs Novarrium, waits three days, downloads a 280-page manuscript, skims it, puts his name on the cover, self-publishes. He writes, let's say, 0 % of his text.

All three, when asked, can say the same sentence: "I use AI to write." The European appeal and the American letter explicitly target writer C. They implicitly question writer B. They say nothing about writer A. And yet in press coverage, the three are routinely collapsed into one.

What the letters actually say

The U.S. letter is precise. It asks publishers: not to release books created by machines; not to use AI tools built on copyrighted content without authors' consent or compensation; not to replace publishing house employees with AI tools; and to only hire human audiobook narrators.

The joint statement from the European Writers Council (the authors), CEATL (literary translators) and FEP (European publishers) asks the European Commission: to require publishers to label AI-generated books; to reserve cultural public funding to human-authored works; and not to recognize AI-generated productions as cultural goods.

Read these lists twice. Neither asks an author to stop using a synonym lookup powered by AI, or to stop asking a model whether Vienna had a tram in 1903. Neither bans tools that help organize a manuscript, suggest a title, or offer a sentence variant on the author's request. What both letters target is the industrial pipeline: models trained without consent on authors' corpora, fully generated books published under fake or real names, synthetic audio narration replacing human voice actors.

Put differently: the letters target substitution, not assistance.

The parallel no one dares to draw — developers

Developers adopted AI in roughly two years, and no one claims they stopped being developers. They argue with the tool. They reject 70 % of what it suggests. They make a judgment call on every line. The AI rephrases, proposes, reorders a function that smelled of sweat. Sometimes — fairly randomly, and without intent, since a model doesn't understand what it writes — it produces a formulation that unlocks a real idea. But the developer decides what enters the code.

Why do we imagine writers would be different? We are also craftspeople in charge. The AI is a sparring partner, not a substitute. It makes us work the thought; it doesn't replace it. Asking a model for a synonym, for a way to reorder a scene, for a Lisbon street name in 1976, is no more cheating than asking Cursor to generate a function signature you're going to review and correct.

The line no one has drawn — "vibe writing"

The predictable objection comes next: fine, but at what point does it become a problem?

It becomes a problem when 80 % of the novel is written entirely by AI, with minimal modification, minimal personal involvement. We can call this vibe writing, by parallel with vibe coding — the practice serious developers despise, which consists of running prompts, accepting whatever comes out, and shipping the code unread. No one in the profession takes a vibe coder seriously. And yet the practice exists, it's spreading, it's doing damage.

On the writing side, the line is the same. A writer who asks a model for a synonym, or who has the beats of an already-written scene reordered, is not a vibe writer. A writer who generates 280 pages from a prompt and signs the cover, is. The distinction is sharp, and it has existed in the dev world for two years now — no reason to reinvent a fuzzier version on the literary side.

Both appeals — the one from the European Writers Council (the European authors) and the one from the 70+ American novelists — target the vibe writer. Not the dialoguing novelist. By collapsing the two, the press gives vibe writers an alibi they don't deserve, and dialoguers a bad conscience they don't deserve either.

The confusion that helps the industry it pretends to fight

Here's what bothers me. Every time the discussion slips from "AI-written books" to "AI in writing in general," two camps benefit.

The vendors of automatic generation benefit because the confusion makes them invisible. If every writing app touching AI is suspect, then no one bothers distinguishing the one that suggests a synonym from the one that writes the chapter for you. Everything turns grey. And in a moral debate, grey is the ally of those with something to hide.

Radical sceptics benefit in the other direction. The confusion lets them wave away any online writing app that contains AI, without bothering to distinguish what it actually does. Faster, morally comfortable, and conveniently avoids the real question.

Meanwhile, the honest writer — the one who wants to write her own book, who has no interest in a machine taking her place, but who would like a well-built tool to organize chapters and suggest a word when she's stuck — is left without a compass.

The two questions that matter

When you evaluate an online writing app — Dabble, NovelCrafter, WriteControl, Extypis, Scribbook, Novlr, any of them — two questions are enough to sort what's compatible with the appeals and what isn't.

1. Do my texts feed a model?

Three possible answers, and this is where I'm going to say something that won't please either camp in the current debate.

First answer: "yes, your texts can be used to train our models, without consent or compensation." That's a violation of the U.S. letter's first point. There's no reason to use it.

Second answer: "yes, on opt-in basis, with compensation." I'm in favor. An author who consciously chooses to license their texts for $2,500 — half of the HarperCollins/Microsoft per-title flat fee — is making a choice that's not illegitimate at all. They're contributing to the making of a model that, tomorrow, will help authors better precisely because it was trained on actual literature, under contract, with explicit consent. The scandal was never the training. The scandal is unconsented, uncompensated, undocumented training. Conflating the two is conflating theft and sale.

Third answer: "no, never — contractual and architectural guarantee." This is the position I took on Extypis today — out of caution, and out of consistency with users who don't all want to be asked the question. WriteControl and Novlr hold it by a different route, by having no AI at all, which is another valid way of not crossing that line. No corpus, no fine-tuning, no silent export. Your texts stay yours.

2. Does the tool take the initiative of writing for me?

If the answer is "yes, the tool can generate an entire chapter from an outline, or rewrite passages without being asked," you're in the vibe writing zone. The book that comes out isn't written by you in the sense the American letter means.

If the answer is "no, the tool stays silent while you write, and only proposes something when you explicitly call for it," the tool is compatible. You remain 100 % the author of your text. The AI holds the door; it doesn't cross the threshold.

On those two criteria, the landscape of online writing apps sorts itself out fairly easily. Sudowrite, Novarrium, Squibler clearly fall on the wrong side of question 2 — they explicitly sell chapter generation as their core value. Dabble, NovelCrafter, Extypis sit on "no to question 2," to varying degrees — AI is present but only on invitation. WriteControl, Novlr, Scribbook don't ask the question because they don't include AI at all. None is intrinsically better than the others; it's about where you want to stand.

Keep the debate precise

The European appeal and the American letter are right. Publishers should pledge not to release generated books. Public funding should go to human works. Models should compensate the authors they're trained on. All of this is fair, and worth defending.

But supporting the appeal doesn't require giving up writing with digital tools that include AI, provided you accept the same discipline developers did: you argue, you decide, you stay in charge. Vibe writing is the red line; everything short of that is still writing.

If you sign the petition — and you should — keep the other eye open: ask your online writing app the two questions. If it answers "your texts never leave our servers to train anything, and our AI only runs when you explicitly ask," you can keep using it without contradiction. If it answers anything else, it might be time to look elsewhere.

Sources:

HU

Hubert Giorgi

Author

Ready to write with the right tools?

Extypis is a free writing workshop built for authors: structure your projects, manage your characters, and export your manuscripts.

Free, no credit card required