AI is homogenizing language and diminishing personal expression

A new study published in Nature Human Behavior suggests that large language models (LLMs) are reducing linguistic diversity by pushing human writing toward common stylistic norms. According to Pew Research Center, over a third of web pages published since ChatGPT’s November 2022 launch were authored by AI, with writing becoming increasingly simpler. Lead researcher Zhivar Sourati, inspired by James Pennebaker’s work on linguistic identity, explored whether LLMs might erode individual expression. His analysis of 80,000 academic papers, 400,000 news articles, and 300,000 Reddit posts revealed that writing complexity converged toward common norms after AI tools became widespread. In experiments, GPT-3.5, Gemini, and Llama 3 reduced writing variation by 21% to 50%. Sourati emphasized the societal implications, asking how LLMs might obscure identity signals in writing.