AI Can Write a Novel. But It Still Can't Write Literature.

The French literary world is in uproar over C'était ça ou mourir (It Was Either That or Die), a debut novel by Canadian-Haitian author Thélyson Orélien. The book won the Prix du Roman Fnac and was longlisted for the Goncourt and the Renaudot - then an anonymous group called “Balance ton Claude” claimed it was “almost totally” written by AI, citing the detection tool Pangram. Orélien denied the accusation outright: “I have always loved creation and literature. I can’t see why I would feel compelled to use a machine”. His publisher, Grasset, defended him, calling the allegations “atrociously injurious” to a writer with an exceptional life story. The row has revived a familiar question with new urgency: if a book can be suspected of being machine-written, what exactly are prizes, publishers, and readers trying to certify?

My take: The longlisting itself isn't infuriating. What's really infuriating is the fact that the industry had to be embarrassed into remembering its purpose.

The withdrawal of C'était ça ou mourir is being sold as a cautionary tale about AI. That is nonsense. It is a market test, and readers failed the machine, not vice versa. Legitimacy vanished alongside the human signature, and this should terrify the publishers treating literature like some glorified content stream.

Of course AI can produce a passable story. We've already seen that it can mimic style, pace, even the most visceral of human emotions. And we all understand that what it cannot do is mean anything. It has no childhood to betray, no grief to transmute, no stubborn hope to risk on a page. This is strongly hypocritical to what literature is in itself, because at the end of the day, literature requires a soul, not just a syntax. Yet a machine has no skin in the game, so its sentences bear no stakes. It cannot bleed the way our authors do.

The elephant in the room is not the age-old, cliché debate over whether machines will replace writers. Avert your gaze instead to the publishers who behave like machines in their desperate pursuit of scale, speed and superficial safety while declaring it "innovation" (which it is not). If your business model needs AI to write books, don't grace yourself with the title of publisher, because all you're doing then is content farming with a literary veneer.

But here is where the hot take gets a second gear. The answer is not a blanket ban on the technology, but it is not acceptance either. The useful question would not be “Did AI touch this?” but “Who is accountable for the result, and do they have the literacy to know what they’re accountable for?” Notice that two kinds of literacy are doing the work here. Foundational literacy is the slow, human capacity to read closely, argue, revise, and recognise a sentence that means something, and it is what lets a writer stay as the responsible author rather than a name pasted onto AI output. Technological literacy is the ability to understand what these models actually do, where they fail, what data they were trained on, and it is what stops us from either worshipping the machine or burning it at the stake. Without both, “AI-assisted” becomes an alibi, and “anti-AI” becomes performative. With both, the question receives a reasonable answer: AI can be legitimate for research, structure-checking, translation drafts, metadata, and admin - the stuff that eats a writer’s time without shaping the voice, but only this and no more.

Literacy should not be a victim to efficiency. Automate the admin, the metadata, the blurbs, whatnot, but please - leave the writing to the people who actually have something to lose.