NYU and UT Austin study identifies six patterns that make language model writing hard to read

Researchers at New York University and the University of Texas at Austin report that frontier language models consistently overuse linguistic patterns that make their output harder to read, according to The Decoder. The team had 14 models write in three settings: casual conversation, coding sessions and argumentative essays. The models included Claude, GPT-5.6, GLM-5, Kimi K2.5 and smaller open-source models such as Llama and Olmo. The researchers analyzed 120 conversations, 45 coding sessions and 980 essays, and compared them with human-written text from movie subtitles, StackExchange and student essays. They identified six patterns. Among them, models lean on unusual word combinations that rarely appear in human writing, such as "reader-patience budget," and they place related parts of a sentence far apart in ways that strain working memory. The study puts numbers on a common complaint: answers can be factually correct and still tiring to read. That affects anyone who relies on model output for work or study, and it gives developers specific targets for improving readability.