If Julius Caesar had debuted this year, William Shakespeare might have been accused of writing it with AI. A certain suspicious rhetorical device appears again and again in the play: “The fault, dear Brutus, is not in our stars, but in ourselves.” “Not that I loved Caesar less, but that I loved Rome more.” “I come to bury Caesar, not to praise him.” These famous lines include what has become perhaps the best-known tic of AI writing - a sentence that tells you what the subject isn’t as well as what it is: It’s not X; it’s Y.
Once you start noticing the construction, you see it all over the place. Citizens Financial Group reported that growth in its private-banking division was “not just a win for the private bank - it’s a win for the entire enterprise.” Michael Flynn wrote, “The target was never a man. The target was the truth.” The horror novel Shy Girl, pulled by its publisher this year, featured lines like “No bag, no things, no armor, just me,” fueling accusations of AI writing. (The author denied using AI. Citizens Financial Group has said its communications team “leverages the technology.” Flynn did not respond to a request for comment.)
The prevalence of this device isn’t just anecdotal - it’s measurable. (Sorry.) Barron’s reported that its appearance in corporate communications more than quadrupled from 2023 to 2025. Researchers at Pangram, an AI-detection tool maker, estimate that Not just X but Y sentences appear three times as often in AI writing as in human writing. Elyas Masrour, a founding engineer at Pangram, told me that all major chatbots - ChatGPT, Claude, Gemini, and various open-source models - rely on it.
Other chatbot tells, like the usage of delve, have come and gone. Last fall, ChatGPT became obsessed with goblins and gremlins, prompting OpenAI to retire its “nerdy” personality. Yet It’s not X; it’s Y has shown no signs of abating. Before ChatGPT, the construction was obscure enough that it didn’t have an agreed-upon name. Now there’s a scramble: terms from academia like antithesis and metalinguistic negation capture some forms but not others. Laurentia Romaniuk, a product manager at OpenAI, calls it “contrastive phrasing.” Despite its clunkiness, the most popular name is “negative parallelism.”
When deployed judiciously, negative parallelism can be punchy. But ChatGPT turns to it too often, Romaniuk acknowledged, so the company is working on broadening the chatbot’s repertoire. In the meantime, users can give ChatGPT “custom instructions.” On a Reddit forum, users trade tips for scrubbing negative parallelism - one suggested pasting Claude’s output into another AI chatbot and telling it to act as a copy editor with a strict ban on “negative pairings.”
One obstacle to a fix is that no one seems to know why AI models are so enamored with negative parallelism - maybe not even the companies that created them. (Anthropic and Google did not respond to interview requests.) The simplest theory is that humans trained them that way. Large language models are built by identifying patterns in human-written text: books, academic papers, patent filings, and the internet. Negative parallelism was present in the data - Shakespeare aside, Vince Lombardi popularized “winning isn’t everything; it’s the only thing,” and DiGiorno’s commercials insisted “It’s not delivery. It’s DiGiorno.”
But the training data also included bad writing that AI companies don’t want their chatbots to mimic, Tuhin Chakrabarty, a computer-science professor at Stony Brook University, told me. So models undergo “reinforcement learning,” where human reviewers grade responses. Chakrabarty said it’s plausible that reviewers gave high marks to responses with It’s not X; it’s Y, because negative parallelism gives the impression of nuance and insight.
That may not explain just how prevalent the construction is. Several experts pointed to another, weirder explanation: Chatbots are text-prediction machines. They generate answers one token at a time, balancing statistical likelihood and high-rated responses. When a chatbot uses negative parallelism, it’s hedging between a clever word choice and an obvious one. Once it starts a sentence characterizing something, the path of least resistance is to say first what the thing isn’t (X), then what it is (Y). “This is” followed by “not just” is both more likely and safer than direct characterization.
Even if researchers could figure out why chatbots embrace negative parallelism, it could be very hard to fix. “When something gets into these models, it’s very hard to pull it out,” Masrour said. Models evolve by training on text generated by other bots, which is replete with negative parallelism. A growing share of internet writing is AI-generated, becoming training data for future models. Some AI labs use AI instead of human reviewers in post-training, risking “model collapse,” where AI reinforces its biases. “It’s a very vicious loop,” Chakrabarty said. “There’s already negative parallelism in the text, and then AI is preferencing negative parallelism - it comes to a point where it just cannot write without that.”
Chatbot clichés might be grating, but there’s an upside: They make AI writing easier to distinguish from human writing. Masrour said Pangram’s software isn’t getting any worse at detection. The trade-off is that a once-potent rhetorical device is now a cliché that makes you sound like a bot. A recent study by researchers in Germany suggested that AI’s writing tics are cropping up in spontaneous human conversation. If that continues, maybe negative parallelism will lose its status as an AI-writing tell after all. The fault, dear readers, will be not in our chatbots, but in ourselves.