"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 more than 70 authors to American publishers, June 2025)
Within a few weeks in the spring of 2025, two coordinated appeals put generative AI on every publisher's agenda. In May, the European Writers' Council (the federation of European authors, representing 220,000 professional writers in 32 countries), together with CEATL (the European council of literary translators' associations) and FEP (the Federation of European Publishers), issued a joint statement asking member states and the European Commission to protect human works, label AI-generated content and reserve public cultural funding for books written by human beings. In June, an open letter signed by more than 70 authors, including Dennis Lehane, Gregory Maguire, Lauren Groff, Jodi Picoult and Paul Tremblay, asked the Big Five American publishers (Penguin Random House, HarperCollins, Simon & Schuster, Hachette and Macmillan) to commit to never publishing books created by machines, and the petition that came with it gathered more than 1,100 signatures in under twenty-four hours.
Their position is clear, and I support it.
One detail still deserves a pause. HarperCollins, one of the publishers the letter addresses, signed a deal with Microsoft in November 2024 licensing its nonfiction backlist at $5,000 per title to train models. A HarperCollins author who signs the petition is therefore asking their own publisher to give up something it already sells. You can sign anyway, with even more reason, while keeping your eyes open to the contradictions of the camp you're joining.
Three authors, the same sentence, three practices
Take three imaginary but very realistic authors.
The first writes crime fiction. He uses ChatGPT to check the spelling of a Basque name he has heard only once, to find out whether a train really ran from Vienna to Trieste in 1903, and to find a synonym when he gets stuck on a word for five minutes. Everything else, the sentence, the rhythm, the silence between two lines of dialogue, the decision to cut a chapter, comes from him, so he writes pretty much all of his text.
The second writes fantasy. He asks Sudowrite to generate a chapter from a scene card, keeps about 70% of it, rewrites the rest by hand, adds his humor and removes whatever rings hollow. He writes perhaps 30% of his text, and the rest is a co-production.
The third doesn't bother. He writes a three-paragraph prompt, launches Novarrium, waits three days, downloads a 280-page manuscript, skims it, puts his name on the cover and self-publishes it, without having written a line of his text.
Ask them, and all three can say they "use AI to write." The European appeal and the American letter explicitly target the third, implicitly question the second and say nothing about the first, while the press regularly lumps all three together.
What the two texts really say
The American letter is precise. It asks publishers not to publish books created by machines, not to use AI tools trained on copyrighted works without their authors' consent and payment, not to replace their staff with AI, and to use only human narrators for audiobooks.
The joint statement from the European Writers' Council (authors), CEATL (literary translators) and FEP (publishers) asks the European Commission to require publishers to label AI-generated books, to reserve public cultural funding for works produced by human beings, and to deny AI output the status of cultural goods.
Neither text asks to forbid a writer from looking up a synonym with an AI's help, or from asking a model whether Vienna really had a tram in 1903. Neither asks to ban tools that help organize a manuscript, suggest a title or propose an alternative sentence when the writer asks. Both appeals target the industrial chain: models trained on authors' works without consent, fully generated books published under real or invented names, synthetic voices replacing actors in audiobooks. What they fight is the replacement of the author, and the help an author chooses to receive falls outside their scope.
The comparison few dare to make: developers
Developers adopted AI within two years, and nobody claims they stopped being developers. They talk with the tool, reject most of what it suggests and decide on every line. The AI rephrases, suggests, tidies up a confused function, and now and then, rather randomly since a model doesn't understand what it writes, it produces a phrasing that unlocks a real idea. It's still the developer who decides what goes into the code.
Nothing forces us to think writers work any differently, because they too run their own building site. The AI serves as a sparring partner that pushes them to sharpen their thinking without ever thinking for them. Asking a model for a synonym, a new order for a scene or a street name in 1976 Lisbon is no more cheating than asking Cursor to write the signature of a function you'll reread and correct.
The line nobody has drawn: "vibe writing"
That leaves the predictable objection: at what point is there a problem?
There's one when 80% of the novel is written by AI, with little editing and little involvement from the author. You can call this vibe writing, by analogy with vibe coding, which serious developers (I'm one of them) look down on, and which consists of sending prompts to the machine, accepting whatever comes out and shipping the code without reading it. No professional takes a vibe coder seriously, which doesn't stop the practice from spreading and doing damage.
For writing, the line runs in the same place. A writer who asks for a synonym, or has the beats of a scene they've already written reordered, remains a writer, whereas one who generates 280 pages from a prompt and signs the cover is vibe writing. Developers have drawn this distinction for two years, and there's no reason to make it blurrier in literature.
Both appeals, the European Writers' Council's and the American novelists', target the vibe writer and leave alone the novelist who talks things through with the tool. By mixing the two, the press gives vibe writers an alibi they don't deserve, and everyone else a guilty conscience they don't deserve either.
A confusion that suits those it claims to fight
Every time the discussion slides from "books written by AI" to "AI in writing in general," two camps benefit.
Sellers of automatic generation benefit because the confusion makes them invisible. If every writing tool that touches AI becomes suspect, nobody can tell the one that suggests a synonym from the one that writes the chapter for you, and that blur always suits those with something to hide.
The most radical skeptics benefit in the other direction, since the confusion lets them dismiss any writing app with AI in it without looking at what it does. It's quick and comfortable, and it spares them from thinking about the real question.
Meanwhile, the honest writer is left without bearings. They want to write their own book and have no wish for a machine to take their place, but they'd welcome a well-made tool to organize their chapters and suggest a word when they're stuck.
The two questions that matter
To judge an online writing app, whether it's Dabble, NovelCrafter, WriteControl, Extypis, Scribbook, Novlr or another, two questions are enough to tell whether it's compatible with these appeals.
1. Are my texts used to train a model?
Three answers are possible, and the second will please nobody in today's debate.
First answer: "yes, your texts may be used to train our models, without your consent or compensation." That's exactly what the first point of the American letter denounces, and you have no reason to use that tool.
Second answer: "yes, if you agree, and for payment." I'm in favor. An author who knowingly licenses their texts for $2,500, half the HarperCollins and Microsoft fee, makes a legitimate choice. They contribute to a model that will help authors better tomorrow because it was trained on real books, under contract and with consent. What's scandalous is training without consent, without payment and without transparency, and confusing it with licensed training amounts to confusing theft and sale.
Third answer: "no, never, and we guarantee it both by contract and by the design of the tool." That's the position I took for Extypis, out of caution and out of respect for users who don't all want to have to ask themselves the question. WriteControl and Novlr get there differently, by including no AI at all, which is another valid way of staying on the right side of that line. At Extypis, no text joins a training corpus, no model is fine-tuned on your writing and nothing is exported behind your back: your texts stay yours.
2. Does the tool take the initiative of writing for me?
If the tool can generate an entire chapter from a card, or rewrite passages without being asked, you're entering vibe writing territory, and the book that comes out is no longer written by you in the sense the American letter means.
If the tool stays silent while you write and only offers something when you call on it, it's compatible with these appeals, and you remain the author of your whole text. The AI opens the door, and you choose whether to walk through it.
With these two criteria, the landscape of online writing apps sorts itself out easily. Sudowrite, Novarrium and Squibler answer the second question badly, since they present chapter generation as their main selling point. Dabble, NovelCrafter and Extypis offer AI, to varying degrees, but only when the writer asks for it. WriteControl, Novlr and Scribbook don't need to ask the question, having no AI. The right tool depends on where you want to stand, and that choice is yours.
Keeping the debate precise
The European appeal and the American letter are right. Publishers should commit not to publish generated books, public funding should go to human works, and companies that train models should pay the authors whose work they use. All of this is fair and deserves defending.
Supporting these appeals doesn't require giving up writing tools that contain AI, as long as you hold yourself to the same discipline as developers: talk with the tool, decide for yourself and stay in charge of your text. Vibe writing marks the red line, and everything on this side of it is still writing.
If you sign the petition, and I encourage you to, also put the two questions to your writing app. If it tells you that your texts never train anything and that its AI only kicks in when you ask, you can keep using it without contradiction; if it tells you something else, it may be time to switch.
Sources:
- European Writers' Council, Joint Statement: A Call for Transparency Regarding AI-Generated Books, May 2025
- Actualitté, Non aux faux livres IA : l'appel des écrivains et éditeurs européens, April 2025
- Literary Hub, Against AI: An Open Letter From Writers to Publishers, June 2025
- NPR, Authors petition publishers to curtail their use of AI, 28 June 2025
- Transparency Coalition, HarperCollins AI Deal with Microsoft Sets First Public Price for Training Data, November 2024
- The Authors Guild, HarperCollins AI Licensing Deal, 2024
Hubert Giorgi
Author
The writing studio novelists were missing
Outline, characters, places and objects, the links between them, mind map, scene board, export: everything lives in one place. Plus an AI that knows your story, whenever you want one.
Free to write and to export. No credit card asked.