The AI Writing Tell 'It's Not X, It's Y' and How Authors Can Avoid It

<img src="/wp-content/uploads/2026/03/avoiding_ai_writing_pitfalls_aq3u2.jpg" alt="Author reviewing a manuscript to remove a common AI writing tic" style="width:100%;border-radius:12px;margin-bottom:2rem;">

A single sentence pattern has become the clearest giveaway that a paragraph was written by AI, and it keeps showing up even in text nobody meant to publish under a machine's byline. According to The Atlantic, in a piece by Will Oremus published on July 12, 2026, the construction "it's not X, it's Y" (also known as negative parallelism) is now the most recognizable tic of AI-generated writing, and it appears across chatbots from every major lab.

Why the pattern is so hard to shake

The Atlantic reports that OpenAI is aware of the problem. Laurentia Romaniuk, a product manager for model behavior at OpenAI, prefers to call the construction "contrastive phrasing" and says the company is working to broaden the range of ways its models express contrast and nuance. Even so, nobody fully understands why chatbots default to this specific pattern so often. The leading theory described in the article is that the phrasing showed up frequently in the human-written text models trained on, and that human reviewers rating model responses during training tended to score sentences built this way higher, because the contrast gives an impression of insight that a flatter sentence does not.

Once a reader starts noticing the pattern, it turns up everywhere, including in writing nobody would guess was AI-assisted. The Atlantic points to a corporate annual report that described growth in a private banking division as more than a win for that division alone, adding that it counted as a win for the entire company too. That additive version, where the second half intensifies or expands on the first rather than truly contradicting it, is just as common as the strict "not X, but Y" form.

What this means if you use AI while writing

For authors who draft with AI tools, this is a useful thing to know rather than a reason to panic. AI-assisted drafting and AI-generated prose are not the same thing, and readers, editors, and now literary prizes are increasingly attuned to phrasing that reads as generic or formulaic. If you use AI to get past a blank page or speed up a first draft, a quick pass to hunt down "it's not X, it's Y" constructions, along with other repeated rhetorical crutches, is a fast way to make sure your own voice comes through in the final version rather than the model's habits.

That kind of editing pass matters more the more AI you use in your process. Keeping your own sentence rhythms, and rewriting any contrastive phrasing that sounds borrowed rather than earned, is one of the simplest ways to keep a manuscript sounding like you wrote it. If you want a writing environment built around exactly that balance, where AI helps you draft faster but you stay the one shaping the voice, you can try a free trial and see how that workflow fits your own writing process.