I’ll start with the conclusion, so if it annoys you, you can spare yourself the rest: how I write is my business. I can use my grandfather’s fountain pen, a typewriter, Word with autocorrect, an editor paid by the page who rewrites half a chapter for me, or a language model that gives me twenty versions of a paragraph, from which I keep three lines. I may have read a thousand books and drawn style and ideas from them, or I may never have read a single one. I can take a plot from a story I overheard in a bar or have one suggested to me in a dream; I can write about real autobiographical events or make them up entirely and claim they happened. Readers have every right to judge what they find on the page and stop reading if they don’t like it. The rest, however, is my business. And if you insist on discovering the truth, know this: it will always remain beyond your reach. You will never have certainty. Writing conceals its author in a thousand different ways, and it is never, ever, and I mean ever, the product of a single pen, even if you lock the writer in solitary confinement.
Now, to the little controversy. On September 21, Thélyson Orélien, a Canadian poet and writer of Haitian origin, won the Prix du roman Fnac for his novel C’était ça ou mourir, which had sold around thirty-five thousand copies in a month. A few hours later, an anonymous account on X called “Balance ton Claude,” which had ninety-two followers, accused him of having written the book “almost entirely” with artificial intelligence, based on the verdict of an AI-detection program called Pangram. It later emerged that Samuel Fitoussi, a right-wing columnist for Le Figaro, was also behind the account, and had immediately amplified the post from his personal profile. Le Monde fed more than half the novel into Pangram and got a result of over 95 percent AI-generated text. France 24 analyzed the entire book and got 92 percent. Meanwhile, articles Orélien published between 2012 and 2014 surfaced in which parts had been copied from other writers. Oh no, COPIED! Off with his head! cries the mob.
Hold on, though. Let’s follow Alberto Puliafito’s reconstruction in Internazionale: “We don’t know under what working conditions the author was writing in 2012, and we should ask ourselves what would happen if we scrutinized the entire daily journalistic output of the whole world. We would probably discover that most of what gets produced is reworked material: there is even an English term for the practice, churnalism. A British study from 2008, for example, found that roughly 80 percent of journalistic content in the UK was not original, and that stories based on original reporting made up just 12 percent.” Why not put the exploitative system behind journalism in the stocks, instead of the 80 percent of journalists who copy?
Back to Orélien. On September 25, the Académie Goncourt removed the book from its selection in order to “preserve the integrity” of the prize. Do we know the truth? No, and it shouldn’t matter. If you want my opinion, I think Orélien did use a language model. But the question I find much more interesting is: why did he lie? And the answer has more to do with you than with him.
Let’s clear up how this trial came about. Alberto Puliafito points out that AFP submitted the passages flagged on X to eight other detectors, with results ranging from “very likely written by a human” to a 100 percent probability of AI generation, with virtually every percentage in between. With results like these, he notes, everyone simply believes whatever they already wanted to believe, although Pangram, it should be said, claims a higher level of accuracy. Shame it isn’t free, because you might get some entertaining surprises if you ran the writing of people who scream themselves hoarse about AI through it. I did. For some people, I got results ranging from 40 to 100 percent AI. Hypocrisy or defective tool? The verdict is yours.
Puliafito also cites a 2025 study in which Shoumik Saha and Soheil Feizi of the University of Maryland showed that even light editing by a language model can cause human-written text to be classified as AI-generated, and that none of the twelve detectors they examined could reliably distinguish a light touch-up from a full rewrite. Pangram says it has improved since then, and in Orélien’s case the cross-checks make its verdict credible. But let’s accept the company’s own stated error rate: one false positive for every twenty-four thousand texts. Puliafito does the math and arrives at roughly forty people falsely accused for every million texts checked. Multiply that million by everything written every day in any given country and you get some sense of the literary Gestapo many people seem to be dreaming of.
If you assume that when a book is written with artificial intelligence the machine must have done everything, you’re wrong. In fact, it’s almost the opposite. If Orélien really did write a good novel using a language model, he pulled off something remarkable. The book won the Prix du roman Fnac, based on the choices of four hundred booksellers and four hundred readers; it made the early selections for the Goncourt, Renaudot, Femina, and Médicis prizes; and publishing rights were sold in around twenty countries. Anyone talking about an absence of work has clearly never tried it.
Luciano Floridi has given this practice a name: distant writing, modeled on Franco Moretti’s distant reading. The author designs the book without physically writing every sentence, establishes constraints, chooses among variants, corrects course, and creativity shifts from drafting to architecture. Writing three hundred pages of fiction with a tool that tends, sentence by sentence, to slide toward the statistical average is not easy. The result is almost always garbage. In my experience, literary writing, unlike nonfiction, is the hardest thing to do with AI, which is precisely why it is not especially common. Using AI as an editor, on the other hand, is very easy, very useful, and therefore extremely widespread.
Luca Sofri pointed out an interesting paradox on his blog. For years we were told that no artificial intelligence would ever match the quality of human talent, especially in art and creative work. Now that a book welcomed by publishers, critics, and prize juries as “a really great book” may turn out to have been written with a machine, suddenly all the credit goes to the software. First AI couldn’t produce anything beautiful; now it apparently did everything by itself.
If the machine alone cannot write anything beautiful, then someone worked very hard on that book and deserves the credit. If the machine can write something beautiful, then we need to abandon the idea that AI cannot create artistic work. My answer, at least for now, is the first. I know this from experience.
But let’s take the accusation seriously: the burden of proof belongs to the person making it. You think it’s easy? You think the computer did everything? That is a serious accusation. You are delegitimizing someone else’s work. Prove it.
Buy a subscription, which costs roughly the price of two pizzas a month, use whatever model you like, and with one or two prompts have it write you a novel capable of winning the Prix Fnac. Or even just this article. Without having my article in front of you, obviously. Without my notes, my reading, or my ideas. If the only way you can get something comparable is by first feeding the machine what I wrote, you have already proved the opposite of what you claim.
If it worked the way you imagine, publishers would be drowning in masterpieces instead of the same old garbage, both human and human-plus-AI. Everyone has access to the same models, yet the results are distributed just as they are with any other technique: an enormous mass of mediocrity and a tiny fringe of successful work. It is what happened to photography when cameras arrived on our phones. What makes the difference is who uses the tool and how. If you think otherwise, there is a nearly free experiment available to prove it, and I do wonder why no one has ever taken me up on the challenge.
Even if we assumed the machine did everything, which, again, is extremely unlikely, you would still have to explain what would actually change for you as a reader. Thirty-five thousand people read that novel in a month, and some of them were moved by it. Amélie Nothomb called it “a prodigious book, a tour de force”; Robert Ménard, the mayor of Béziers, whose relationship with welcoming migrants is notoriously chilly, said that no book had shaken him this deeply in years, to the point of changing the way he saw migrants themselves.
Those emotions happened.
Discovering how the text was produced may alter your judgment afterward while leaving intact what the act of reading already did to you. Calvino understood this as early as 1967 in Cybernetics and Ghosts, where he imagined a machine capable of generating every possible permutation of a given body of material and observed that meaning would still arise from the encounter between one of those combinations and a human consciousness, with its history and its unconscious. When you think about it, the body truly at stake in a novel is first and foremost the reader’s.
There is also an almost comic hypocrisy here. When a system from the same technological family helps predict how a protein folds, you applaud the result and don’t demand certificates of purity from the laboratory. When it touches a novel, excommunication begins. In science, you judge the outcome. In the arts, you demand a pedigree.
The purity you are asking for, in any case, has never existed in literature.
In 1592, a pamphlet attributed to Robert Greene warned playwrights about an “upstart crow, beautified with our feathers,” and the crow was almost certainly Shakespeare, who really did take feathers wherever he found them: the plot of Romeo and Juliet from Arthur Brooke’s poem; English history from Holinshed’s Chronicles; Cleopatra on her barge in Antony and Cleopatra almost word for word from Plutarch as translated by Thomas North; Gonzalo’s utopia in The Tempest from Montaigne as translated by John Florio.
Montaigne himself said that he quoted others only to express himself better, and admitted that he sometimes deliberately hid their names so that hurried critics might slap Plutarch on the nose instead. Dumas put his name on novels to which Auguste Maquet had made substantial contributions. Lautréamont wrote in Poésies that plagiarism is necessary because progress implies it. T. S. Eliot turned that idea, in 1920, into the famous maxim that immature poets imitate and mature poets steal. The literature you worship was built this way too, and nobody has ever removed Antony and Cleopatra from the shelves because Thomas North is inside it.
The obsession with purity also comes at a cost, and honest people are the ones who pay it. Manav Raj, Justin Berg, and Rob Seamans conducted sixteen preregistered experiments involving more than twenty-seven thousand people, published this year in the Journal of Experimental Psychology: General, to see how judgments of creative writing changed when readers were told that a text had been written by AI or with AI assistance.
The result was a consistent penalty averaging around six percent, almost entirely linked to perceived authenticity and resistant to every attempt to mitigate it. Describing the text as a collaboration between human and machine changed nothing. In one of the studies, the researchers used human-written stories that had won a literary competition, and even those were penalized by the same amount once they were labeled as AI-assisted.
The penalty comes from the label, regardless of the text.
You could have written the most beautiful page of the century. All you have to do is disclose the AI assistance and your score drops.
Outside literature, the same pattern appears. In four experiments involving 4,439 participants, Jessica Reif, Richard Larrick, and Jack Soll showed that people who use AI at work expect to be judged more negatively in terms of competence and motivation, and those judgments duly arrive. In an experiment by Atlassian’s Teamwork Lab involving 961 workers, people who said they had used AI assistance were judged ten times lazier and were 24 percentage points less likely to be recommended for a project, even when the work they submitted was identical.
The outcome is predictable: according to a study by KPMG and the University of Melbourne involving forty-eight thousand people across forty-seven countries, 57 percent of employees admit to hiding their use of AI and presenting AI-generated work as entirely their own. As early as 2024, 52 percent of people using these tools at work were reluctant to admit doing so on their most important tasks.
Everyone uses it, and almost no one says so.
The problem, you might say, is the lie. If Orélien had said, “Yes, I used AI too, but the ideas and experiences are mine,” then this would be a different conversation.
The problem is that if he had said that, the novel would never have been published.
And you know what? I think he was right to lie.
If you say, “I want to know, so I can pay less for the book or refuse to read it,” then you are practically begging to be deceived. You demand transparency and punish the people who practice it. As far as the novel itself is concerned, Orélien’s only mistake, in my view, was lying badly, assuming he lied at all.
Raj, Berg, and Seamans themselves warn that legally mandating disclosure may only entrench the prejudice. If you build a factory for liars and then act scandalized by the lies, you deserve the lies you get.
Then there is racism, something often ignored by the left when it joins these controversies born on the right, a right that has no problem using AI itself.
This summer, Jerry Falade, a Nigerian debut novelist and PhD student in Dallas, lost a $2.4 million publishing contract over suspicions that were never substantiated. He pointed out that he was the third Black author that year to have a major contract canceled or frozen because of accusations of the same kind. H. M. Wolfe, the African American author of Daggermouth, had 60 percent of her text classified as AI-generated by Pangram in a study later covered by The Atlantic. Trinidadian writer Jamir Nazir was flagged for allegedly submitting an AI-written story to the Commonwealth Short Story Prize, before the foundation examined his drafts and notes, cleared him, and ultimately declared him the winner.
Now it is a Haitian writer’s turn.
As early as 2023, research from Stanford showed that AI detectors systematically penalize people writing in a language that is not their native one. Suspicion slips more easily into the grooves of racial prejudice, and the names ending up in the stocks tend to resemble one another with a regularity I find difficult to dismiss as coincidence.
Matthew Kirschenbaum recounts in Track Changes that in the late 1970s some novelists concealed the fact that they wrote on computers, and some would retype their entire manuscript on a typewriter before submitting it because a manuscript that looked too clean betrayed the use of technology and raised suspicions that insufficient effort had gone into the work.
A word processor does not generate the text, and the comparison has its limits. I know. Still, that panic disappeared as soon as everyone began using the technology, and today it would never occur to you to ask a novelist whether they wrote their book in Word.
And what about effort? Why am I so hostile to the glorification of effort? Why do I think defending it can be classist?
Let me ask you this: did you choose the language you were born into? Your social class? Whether you had the time and money required for an education? And if we insist on believing in the fantasy of innate genius, you didn’t choose to be born with that either. It happened to you.
I am sure you have worked hard. But you did not choose your predisposition or your capacity for hard work either.
Writing requires free time, “a room of one’s own,” and someone dealing with everything else while you do it. Those who could afford these things have always had an easy time passing privilege off as merit. That is where the snobbery of effort comes from.
As for sexism, you need only look at who actually did much of the work behind certain masterpieces. The myth of the solitary author sweating over the page has always erased the labor of the people around him, very often women.
I’ll repeat it: it is no coincidence that contempt for AI hits the lower social classes hardest, which is why it is disappointing to see so much of the left riding this wave.
And besides, if someone truly could produce a masterpiece without pain and without effort, with machines or without them, we should be delighted.
According to every AI detector I used except Pangram, this text is 100 percent human. According to Pangram, it is 72 percent human. With numbers like those, I would hardly be stoned to death, and yet I used AI frequently and enthusiastically while writing it. The truth is that it is 100 percent mine: the responsibility is mine, the ideas are mine, and the choices are mine. In other texts where I used AI almost exclusively to translate my own writing, Pangram says they are 100 percent AI-generated.
Francesco D’Isa