TechCrunch: The sentence construction that marks AI-generated writing

A Barron’s investigation, covered by TechCrunch on April 20, 2026, found that one sentence construction has become so common in AI-assisted writing that its presence is now a statistical signal in institutional text: “It’s not just X — it’s Y.”

The Barron’s team used AlphaSense to scan corporate press releases, earnings reports, and government filings, and found that this construction appears at measurably higher rates in documents that exhibit other AI-generation signals. Max Spero, CEO of AI detection company Pangram, described the pattern as a stylistic tic preferred by 2025-era frontier language models. He noted that its presence is not proof of AI authorship — the base rate in human writing is non-trivial — but its frequency in formal institutional documents is high enough to be a meaningful signal in statistical detection.

The construction is an emphatic framing device that AI models learned from training data weighted toward persuasive writing. Models use it to signal a contrast or escalation: the first clause introduces a familiar framing, and the second positions the subject as something more significant. Human writers use it too, but with variation in placement and frequency. AI-generated text tends to use it at paragraph ends and in opening lines of press releases, at rates that stand out in corpus analysis.

For writers, the finding has two practical implications. First, editing for these detectable patterns before submission — not to deceive detection tools, but because the patterns typically indicate a structure that is doing rhetorical work without analytical work beneath it. Second, understanding that detection tools are improving in exactly this direction: corpus-level pattern analysis rather than sentence-level fluency checks. A document that reads well can still carry statistical fingerprints.

For communications teams and editorial staff working with AI-assisted drafts, the article is a reminder that review processes need to check for structural patterns, not just factual accuracy and tone. The most common AI writing tells in 2026 are not grammatical errors — they are overused constructions that human writers vary instinctively.