A huge thank you to my amazing co-authors for their contributions to this work:
Farzan Karimi-Malekabadi, Meltem Ozcan, Colin McDaniel, Alireza Ziabari, Jackson Trager, Ala Tak, Meng Chen, Fred Morstatter, and @mdehghani.bsky.social . Excited to share these findings with the community!

These findings raise critical questions: How does AI-assisted writing influence hiring, education, and cultural expression? As LLMs standardize language, they may also shape social and professional opportunities in ways we don’t yet fully understand.

But the impact goes beyond style. LLMs also weaken the connection between language and personal traits, subtly shifting how authors are perceived. This challenges long-established linguistic markers used in areas like personality assessment and demographic inference.

We analyzed writings across multiple platforms and found a measurable decline in the variance of linguistic complexity after ChatGPT’s introduction. Controlled experiments confirmed that LLMs consistently homogenize writing styles, reinforcing this trend at scale.

How do LLMs shape writing? Our research shows they make language more uniform, reducing diversity and altering how personal traits appear in text. These shifts have big implications for identity, culture, and fairness.
#LLMs #NLProc
arxiv.org/abs/2502.11266