Researchers find six patterns that make AI text so exhausting to read

What happened

Researchers at New York University and the University of Texas at Austin identified six linguistic patterns that language models overuse compared with human writing. Their analysis covered 14 models, 120 conversations, 45 coding sessions, and 980 essays, finding constructions such as stacked nouns and widely separated sentence components. Comparisons within the Olmo family suggest post-training may contribute to these patterns, while Claude Opus 5.5 improved over Opus 5 in 19 out of 21 comparisons.

Read next

• More changes in today’s briefingNvidia open-sources Flora with projected robot video labeling costs up to 125x lower than GPT-6 Astra

The story so far

Researchers at New York University and the University of Texas at Austin identified six linguistic patterns that language models overuse compared with human writers and linked them to harder-to-read AI text. Their analysis covered 14 models across 120 conversations, 45 coding sessions, and 980 essays. The patterns included stacked nouns and sentence components placed far apart. Comparisons within the Olmo model family suggested that post-training may contribute to these patterns. The researchers also found that Claude Opus 5.5 improved over Opus 5 in 19 of 21 comparisons, indicating that the patterns varied between model versions.