In early May, the Commonwealth Foundation announced the five regional winners for its influential Short Story Prize, which recognizes unpublished short fiction. One of the awardees, a Trinidadian writer named Jamir Nazir, was accused of A.I.-assisted cheating by a broad array of social-media users who seized upon his story’s synthetic tics, glitchy metaphors, and general unreadability. (“They called her Zoongie,” one passage from the story goes. “Maybe it was a name; maybe rain took a shape and decided to keep it.”) In a statement, Razmi Farook, the director-general of the Commonwealth Foundation, said that contestants had confirmed to the Foundation that they had not used A.I., and that the authors of the short-listed stories had made this attestation twice. The next day, on a call with the Times, Farook allowed that the moment had perhaps come to “look at ourselves internally to see if we feel that our process to date has been robust enough.”
Shortly after Nazir’s story, “The Serpent in the Grove,” appeared online, in the British magazine Granta, Ethan Mollick, a professor at the Wharton School, ran it through the A.I.-detection platform Pangram, which flagged one hundred per cent of the text as likely to have been A.I.-generated. Two of the other winning entries, by the Maltese author John Edward DeMicoli and the Indian author Sharon Aruparayil, were similarly implicated. (Aruparayil denied using A.I. to write her story “Mehendi Nights,” calling the allegations “an entertaining witch-hunt.”) In an interview with the Observer responding to the scandal, Nazir said that his writing process consists largely of speech-to-text dictation on an Android phone. (He cited chronic-health conditions that make sustained typing impossible, and he’s published at least one poem about neuropathy to his Facebook page.) The publisher of Granta, Sigrid Rausing, put out a statement noting that the team had asked the A.I. program Claude about the provenance of “The Serpent in the Grove” but couldn’t say for sure whether “the judges have now awarded a prize to an instance of A.I. plagiarism—we don’t yet know, and perhaps we never will know.”
Epistemically, there is something a bit wobbly about using chatbots to determine whether a piece of prose was written by chatbots. A Stanford study found, in 2023, that A.I.-detecting algorithms tend to be biased against non-native English speakers. Still, as Mollick put it on Bluesky, “Come on, if you know you know.” According to internet lore, A.I.-generated writing can be recognized by a handful of tells, memorably enumerated by Sam Kriss in the Times magazine in 2025. These include anaphora, when words repeat at the beginning of successive clauses, and epistrophe, when words repeat at the end of successive clauses. Nazir’s piece features both. “No fan, no bulb, no hum,” one line begins. “Bush kept it, snakes liked it,” another starts. And: “Water took her and would not return her.”
Then there is zeugma, when a verb takes two objects, one literal and one figurative. Throughout the story, air is “sweet with cane and forgetting” and the mouth of a well is “boarded with ply and chance,” as if a magnet is pulling the sentence away from material reality. Finally, negative parallelism, the “not x but y” construction, which is much reviled by human L.L.M. detectors for its ubiquity in A.I.-generated prose, is all over “The Serpent in the Grove”: laughter is said to “cut a hush, not cure it,” and Nazir writes that “bush took him in—not like a mother, like a judge.” These rhetorical devices exploit our learned associations between certain types of repetition in prose and heightened meaning; they also create a pulse we feel in our bodies. If they recur in automated text, it’s because they recur in human writing, but in the fake stuff they’re decoupled from content.
