The phrase "it's not just X — it's Y" has become an infallible marker of AI-generated texts

There's a simple way to check a text for AI origin: find the construction "it's not just X — it's also Y". If it appears even once, the probability of synthetic authorship approaches one hundred percent. This is the conclusion that follows from observations that TechCrunch summarized in April 2026.

The construction with the opposition "not only... but also..." has become such a characteristic feature of AI-generated materials that it has ceased to be merely circumstantial evidence — now it is almost a guarantee.

Why exactly this one? Language models were trained on billions of texts, among which were journalistic long-reads, analytical reviews, and marketing materials. In these genres, the construction appears frequently — it creates an appearance of depth and nuance.

The model has learned it as a signal of "intelligent writing" and now reproduces it automatically whenever it needs to strengthen an argument or show that the topic has been examined from all sides. The problem is dosage. In human text, such a construction appears rarely and in the right place.

When a language model uses it several times in a row in one material, it is no longer style — it is a glitch that immediately catches the eye of any editor. This is not the first nor the last "clue" of this kind. Before this, researchers have documented other AI markers: the word "delve" in inappropriate contexts, introductory phrases like "Certainly!"

and "Great question!", perfectly even lists of exactly five items, conclusions in the spirit of "this opens new horizons for future research". Each such pattern goes through the same trajectory: first editors notice it, then it falls into the list of detectors, then developers try to remove it — and the cycle repeats.

The scale of the problem is significant. By several estimates, in some niches — marketing blogs, corporate news, content farms — the share of AI-generated materials already exceeds 60-70%. Characteristic patterns have literally spread across the internet, and readers are beginning to recognize them intuitively, even without being able to articulate what exactly bothers them.

For companies and authors who use language models in their work, this is a practical signal on several levels. First: final editing is not optional. Running a text through a model and publishing it immediately means risking your reputation.

A reader who has noticed a formulaic construction draws a conclusion not only about the quality of a particular material, but also about the author's attitude toward the audience. Second: prompt tuning works. An explicit ban on clichés in a system prompt — "don't use construction X", "avoid symmetrical oppositions" — radically reduces their frequency.

This is not a guarantee, but a significant improvement. Third: the list of markers needs to be updated. Patterns change along with models.

What worked as a detector a year ago may have already been corrected in new versions. And conversely — new models bring new clichés that no one has noticed yet. The story with the "it's not just X — it's Y" construction illustrates well the nature of the race between language models and those who edit them.

Models quickly master stylistic techniques that look convincing at a statistical level — but that is precisely why they reproduce them too often and too predictably. A good editor still beats the algorithm. For now.