AI writing assistants are homogenizing language, cutting style diversity 21-50%
Polishing prose with an LLM doesn't strip the facts, but it does strip the fingerprints. A Nature Human Behaviour analysis of 880,000+ texts across seven datasets finds LLM-assisted writing cut style-complexity variance by 21-50%, flattening distinctive voices toward dominant norms. The same homogenization shows up in Reddit posts, academic papers, news and congressional records, with knock-on effects for personality assessment, hiring and cultural preservation.