Research · Updated 24 Oct, 11:34 pm IST

Study finds AI-generated text harder to process than human writing

Why it matters for readers: It helps you understand why some AI-generated writing feels strange and what researchers are changing to fix it.

AI models use psycholinguistic features associated with reading difficulty more often than round-trip translation paraphrases.1
Compared with human writing, frontier and near-frontier models use more inanimate subjects with animate verbs and more low-frequency n-grams.1
Most models also show longer syntactic dependencies, more stacked nouns, and more reduced object relatives in conversation.1
Researchers propose models are optimized for information density (math, code, reasoning) rather than communicative success, explaining the divergence from human-like language.1
A proof-of-concept fine-tune with a length penalty causes models to diverge from human-like linguistic behavior, supporting the hypothesis.1

Get a brief like this every morning

Uzha reads hundreds of sources and gives you the stories that matter for your work, with every source linked. Free.