New York University and University of Texas researchers identify six linguistic habits that make AI output difficult to read
Frontier language models systematically overuse specific patterns that increase the effort required to process their text.
Users of ChatGPT, Claude, or similar systems often note that while answers are factually correct, the reading experience feels laborious. The text is frequently packed with nested clauses, awkward phrasing, and metaphors that do not occur in natural human speech.
A new study from New York University and the University of Texas at Austin provides data to support this observation. The team instructed 14 different models to generate text during casual conversation, coding tasks, and argumentative essays. The group included Claude, GPT-5.6, GLM-5, Kimi K2.5, and smaller open-source versions like Llama and Olmo. Researchers analysed 120 conversations, 45 coding sessions, and 980 essays. They compared these outputs against human-written material sourced from movie subtitles, StackExchange discussions, and student essays.
Six patterns that make AI text a chore to read
AI models frequently employ unusual word combinations that rarely appear in human writing, such as “reader-patience budget.” They also separate related parts of a sentence by long distances, which taxes working memory. For example, Claude Opus 5 generated the sentence “A cat who can always get up and away is a much calmer cat,” where the subject and the main predicate are separated by a lengthy clause.
Shortened references with missing connectors are common. Instead of writing “the book that the professor wrote,” models often produce “the book the professor wrote.” This saves a single word but creates brief confusion regarding how the parts fit together. Stacked nouns like “horror movie soundtrack generator” force readers to determine on their own which word modifies which.
The models turn nouns into verbs where humans would not. Claude Fable 5 wrote “calendar it,” whereas a human would say “put it in the calendar.” They apply animate verbs to inanimate subjects as well. Claude Opus 5 produced “The whole conversation is eating its own tail,” even though a conversation does not have a tail and cannot eat. Readers can infer the meaning, but it requires an extra mental beat.
Smaller models write more like humans
Llama and the Olmo base model come closest to human text. Within the Olmo family, scores rise steadily from the base model through the instruct variant to the reasoning model Olmo Think, which hits frontier-level numbers. This points to post-training as the likely cause. Claude Opus 5.5 beats its predecessor Opus 5 in 19 out of 21 comparisons, suggesting that targeted finetuning can improve readability.
Training rewards information density over clarity
Human language balances ease of production with ease of understanding, but modern language models are trained increasingly on tasks like math, code, and logical reasoning, where correctness gets rewarded and clarity does not, the researchers say.
The shift from RLHF (Reinforcement Learning from Human Feedback) to RLVR (Reinforcement Learning with Verifiable Rewards) removes humans from the feedback loop entirely, the researchers argue. Models develop a kind of “dialect” that works well for compressing information and solving complex problems but poorly for human readers.
As a proof of concept, the researchers trained a smaller language model with a penalty term that rewarded shorter answers. Under certain training conditions, the model favoured shorter phrasings almost without exception, regardless of context, and even generalised that preference to sentence patterns it had never encountered during training. Humans, by contrast, chose between shorter and longer variants depending on the situation.
A similar explanation comes from inside the industry. Anthropic employee Jackson Kernion, who works on Claude fine-tuning, recently explained why Opus 4.6 was the last good “writing model” from Anthropic. Optimising newer models for math, code, and technical explanations aimed at other AI models pushed Claude toward a style that works well for AI but reads as too dense and unnatural for humans. Kernion called it an adaptation to “LLM psychology.” With Opus 5.5, Anthropic pushed back for the first time, addressing sentence clarity and overloaded blocks of information, and as the study shows, successfully so.
