Which AI models were tested in the study?

The study included Claude, GPT-5.6, GLM-5, Kimi K2.5, and smaller open-source models like Llama and Olmo.

Which models wrote most like humans?

Llama and the Olmo base model came closest to human text, while within the Olmo family, scores rose from base to instruct to the reasoning model Olmo Think.

Can finetuning make AI writing more readable?

Yes, Claude Opus 5.5 beat its predecessor Opus 5 in 19 out of 21 comparisons, suggesting targeted finetuning can improve readability.

Large Language ModelsAI Safety & AlignmentTHE DECODER4h agoOct 27, 2026, 04:00 JST

NYU, UT Austin study finds six AI writing patterns that exhaust readers

Send on LINE

3 Key Points

What happened
Researchers at New York University and the University of Texas at Austin had 14 language models, including Claude and GPT-5.6, produce 120 conversations, 45 coding sessions, and 980 essays, then compared them with human text. They identified six overused constructions, such as stacked nouns and animate verbs on inanimate subjects.
Why it matters
These patterns—like turning nouns into verbs and separating related sentence parts—force readers to spend extra effort parsing meaning, which may make AI-generated text feel more taxing than human writing.

WHO IT HITSWriters and editors who rely on AI-generated drafts may need to watch for these six constructions to keep text clear, and AI developers could use the findings to finetune models for more natural readability.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Get these free every morning

Context & Analysis

The study compared AI-generated text against human-written text from movie subtitles, StackExchange, and student essays, finding that models consistently overuse constructions that slow readers down. For example, Claude Opus 5 produced sentences like "A cat who can always get up and away is a much calmer cat," where the subject and predicate are separated by a long clause, taxing working memory. Models also turn nouns into verbs where humans wouldn't, as in Claude Fable 5's "calendar it." The researchers argue that modern language models are trained on tasks like math, code, and logical reasoning, where correctness is rewarded and clarity isn't. Anthropic employee Jackson Kernion, who works on Claude fine-tuning, recently explained that optimizing newer models for math and technical explanations pushed Claude toward a style that reads as too dense for humans. With Opus 5.5, Anthropic addressed sentence clarity and overloaded information blocks, and the study shows it succeeded.

FAQ

Which AI models were tested in the study?: The study included Claude, GPT-5.6, GLM-5, Kimi K2.5, and smaller open-source models like Llama and Olmo.
Which models wrote most like humans?: Llama and the Olmo base model came closest to human text, while within the Olmo family, scores rose from base to instruct to the reasoning model Olmo Think.
Can finetuning make AI writing more readable?: Yes, Claude Opus 5.5 beat its predecessor Opus 5 in 19 out of 21 comparisons, suggesting targeted finetuning can improve readability.

THE DECODERRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.