The most persistent tell of AI-generated text is a rhetorical device that was once a hallmark of human eloquence: the "it's not X, it's Y" construction. Its use in corporate communications more than quadrupled between 2023 and 2025, according to Barron's, and researchers at the AI-detection company Pangram estimate it appears three times as often in chatbot output as in human writing. For writers, the pattern has become so pervasive that it threatens to strip a once-powerful technique of its impact.
From Shakespeare to chatbots
The structure is ancient. Shakespeare used it in "Julius Caesar" - "The fault, dear Brutus, is not in our stars, but in ourselves" - and it surfaces everywhere from Vince Lombardi's "Winning isn't everything; it's the only thing" to DiGiorno's "It's not delivery." But the current flood is different. A former Trump adviser wrote on X that "the target was never a man. The target was the truth." A Citizens Financial Group annual review framed a division's growth as "not just a win for the private bank - it's a win for the entire enterprise." The construction now functions as a shibboleth: readers who spot it often suspect the text was written by a machine.
Elyas Masrour, a founding engineer at Pangram, said all major chatbots - ChatGPT, Claude, Gemini, and open-source models - lean on it to varying degrees. Laurentia Romaniuk, a product manager for model behavior at OpenAI, called it "contrastive phrasing." The more common label among observers is negative parallelism.
Why the tic won't fade
Many other chatbot tells have come and gone. Last fall, OpenAI had to retire ChatGPT's "nerdy" personality after it became obsessed with goblins and gremlins. Negative parallelism, however, shows no signs of abating. One reason may be that human reviewers in the reinforcement-learning process tend to reward it. Tuhin Chakrabarty, a computer-science professor at Stony Brook University, told The Atlantic that responses containing "it's not X, it's Y" can give the impression of nuance and insight - the AI appears to be reasoning toward a better description. That feedback loop would embed the pattern deeper.
But several experts pointed to a stranger explanation. Chatbots are still text-prediction machines that generate one token at a time. When a model starts a sentence meant to characterize something, the path of least resistance is to say what the thing isn't before saying what it is. "This is not just" is both safer and more probable than the many direct openings a human might choose. The structure also reduces the cognitive load on the model for the rest of the sentence. Generative AI and LLM Courses often explore this tension between clever and obvious word choices that shapes all chatbot output.
A feedback loop that eats itself
Once a pattern enters a model, it is hard to remove, Masrour said. Newer models are increasingly trained on text generated by other bots - text already saturated with negative parallelism. At the same time, some AI labs are using AI reviewers in the post-training process, which Chakrabarty warned can lead to "model collapse," where the machine reinforces its own biases until it loses touch with human grounding. "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."
On Reddit forums, users swap tips for scrubbing the tic from chatbot drafts, such as pasting output into a second AI and instructing it to ban "negative pairings." Romaniuk said OpenAI is working to broaden ChatGPT's repertoire, and users can try giving custom instructions. But the sheer scale of AI-generated text now feeding into the internet means the construction is likely being baked into future models as well.
Why this matters for writers
The stubbornness of negative parallelism is a double-edged sword. It makes AI writing easier to detect, but it also turns a legitimate rhetorical tool into a cliché that marks any writer who uses it as a possible bot. A recent study by German researchers suggested AI writing tics are now appearing more often in spontaneous human conversation, which means writers may soon find themselves inadvertently mimicking machine speech. AI for Writers Courses can help professionals recognize these patterns and decide when to avoid them - not by abandoning contrast entirely, but by varying sentence structures so the prose doesn't sound formulaic. For the writer who wants to keep the power of antithesis without the taint of automation, the challenge is to use the device sparingly and to know when the machine is doing the thinking for you.
Your membership also unlocks:
