A few days ago, I attended a conference where the speaker read his talk off his note cards. I immediately recognized ChatGPT’s fingerprints: “and this is where it gets interesting,” “it’s not X, it’s Y”… The topic was management, a field that already generates plenty of hot air without any help from machines. With AI in the mix, I tuned out after a couple of minutes. Had he spoken from an outline in his head, he would almost certainly have been livelier and easier to follow.
Afterward, I realized the problem runs deeper than one dull talk. AI does remove the barrier of linguistic insecurity that holds so many people back, and that matters. But it also spreads a poison through language that contaminates what we produce and, above all, how we receive what others produce. Every email, report and book now arrives with a question attached: how much of it was actually written by a human?
The scandal gripping French literature
In France, that question has just derailed the career of one of this year’s most celebrated novelists. The story has been front-page news all week. On September 21, Thélyson Orélien, a Haitian-born Canadian writer, won the Prix du roman Fnac, a major award run by the French retail chain, for his novel C’était ça ou mourir (“It Was That or Die”). The book follows a Haitian man’s migration route across the Americas to Canada. It was first published in Montreal in March, and is shortlisted in France for nearly every prestigious literary prize, including the Goncourt. Translation offers have come in from about 20 countries.
That same day, an anonymous and little-followed account on X called “Balance ton Claude” (a play on #BalanceTonPorc, the French version of #MeToo) claimed the novel had been written almost entirely by AI, posting screenshots from the AI detection tool Pangram as evidence. The post passed three million views in just two days.
Orélien flatly denied it, saying his imagery and rhythms come from Haitian and Caribbean traditions, and suggesting the software is calibrated on classical French literature. He also denounced a wave of racist attacks. His publisher expressed their “absolute confidence” in him. Then the media ran their own tests, and the results did not help his case. Quebec’s daily La Presse submitted the entire novel to Pangram and got a score of 94%. Other outlets got 100% on every excerpt, while two texts Orélien published in 2014 and 2015, as well as new novels by other French writers, came out human. Yet a detector proves nothing about how a text was actually written, and press agency AFP ran the same passages through eight different detectors, some of which judged them very likely human.
Detectors, humanizers and the endless arms race
Behind the controversy lies an arms race that nobody seems able to win. On one side, detection companies like Pangram, GPTZero and Turnitin train classifiers on each new generation of AI-generated text. On the other, an industry of “humanizers” promises to rewrite chatbot output so it avoids detection, while AI companies themselves keep sanding down their models’ verbal tics. “It’s a game of cat and mouse,” said a French computational linguist, since detectors have to be retrained on models that change every month. Every advance on one side invites a response from the other. And reliability varies by language. Pangram acknowledges that its tool is less reliable on short excerpts, which is exactly what circulated on X.
The humans caught in the middle pay the price. Researchers at Stanford showed in 2023 that earlier detectors disproportionately flagged essays by non-native English speakers as machine-written, because simpler, more predictable phrasing reads as “AI-like.” Orélien’s defenders make a similar argument about his creolized French, full of oral rhythms and aphorisms. Meanwhile, writers have started policing themselves, deleting em dashes, tidy lists of three and anything else that has become a known “AI tell.” Machines are trained to sound more human. Humans edit themselves to sound less like machines. And telling them apart gets harder. So suspicion grows.
Who owns a style?
Many writers came to Orélien’s defense early on. Bestselling novelist Virginie Grimaldi said on Instagram that some of her own texts had been flagged as “written with the help of AI.” She summed up the absurdity well: AI devours writers’ styles and then accuses them of using AI, like a baker who steals your brioche recipe and then accuses you of copying him. Virginie Despentes made much the same point on French television (ending with “Go f… yourself, Pangram !”). AI companies harvested human authors’ styles without permission, and now their tools dispute those authors’ claim to their own voice.
I haven’t read Orélien’s novel, so I have no view on the text itself. But I can see how quickly the accusation became a pretext. Racist comments appeared within hours, and part of the French right treated the suspicion as proof, tying it to the novel’s subject of migration. Some critics had already faulted the book’s repeated aphorisms and its “Not… Not… Just…” constructions before someone ran it through Pangram.
Distrust everywhere, including at the office
Whatever the actual share of AI in what we read, the poison has induced a general distrust. We scrutinize our colleagues’ emails (“he couldn’t even be bothered to use his own words”), the reports and white papers we’re asked to review, and even the paper books we buy precisely to get away from screens. Any turn of phrase ends up sounding like a cliché. Sheer volume, repetition and zero-cost reuse drain language of its lifeblood, as if a vampire had drained it. A text that cost nothing to write makes nobody want to read it. The quantity of text keeps exploding while our attention stays fixed, or rather crumbles as the pile grows. We are heading toward a world saturated with messages and short on people to receive them.
Worse, the contamination reaches texts that were most certainly written without AI, including everything published before 2022. The mannerisms that trained the models now irritate me when I find them in the human authors they came from. I can force my attention back to the page. But it saddens me that no register has been spared, whether casual, formal or literary.
Learning to receive again
Usually I like to offer solutions. This time I’m at a loss. Perhaps silence is the only space the poison hasn’t yet reached. One thing is certain: every piece of writing or speech becomes harder to receive as it becomes easier to produce. What is scarce and precious now is our attention and our ability to read and listen without suspicion. That is probably where we have to start, including with ourselves.
