New research published in Nature Human Behaviour shows that AI writing assistants are flattening the style of written work. A study of 880,000 texts found that after ChatGPT's release, writing became noticeably less varied as AI use spread. The models preserved meaning but compressed the range of language and shifted the personal traits readers infer from the words.
The findings draw on real-world text samples analyzed before and after ChatGPT launched. Researchers used a zero-shot detector to estimate how much AI-generated content appeared in the wild, then compared writing style across that large corpus. The pattern was consistent: as AI adoption grew, stylistic diversity fell.
What the data shows
The study, published August 24, 2026, examined how LLM-assisted writing affects both surface style and deeper signals. Word choice, sentence rhythm, and the small function words that carry psychological weight all became more uniform. Those function words - pronouns, prepositions, articles - are exactly the features that researchers use to infer personality, social background, and emotional state from text.
The authors also found that AI rewriting biased the personal traits that can be read from a person's writing. In other words, the words people choose after editing with an AI assistant may not reflect who they are. The text gets cleaner, but it also gets less personal.
What this means for writers
For people who write professionally, the stakes are direct. The same tools that save time are also erasing the stylistic fingerprints that make individual voices recognizable. A writer who runs every draft through an AI assistant may produce more polished copy - but that copy will sound more like everyone else's polished copy.
The research builds on earlier work showing that language reveals moral concerns and psychological states through everyday word use. It also connects to findings that machine-learning models amplify biases in their training data, and that LLMs can flatten identity groups when they replace human participants. The new study extends that pattern to the act of writing itself.
For editors and hiring managers, the implication is practical: writing samples become less reliable as signals of a candidate's natural voice. For writers, the takeaway is more personal. If your voice is part of what you sell, relying on AI to smooth every sentence may cost you more than it saves.
None of this means abandoning AI tools. It means using them selectively - for research, for structure, for catching errors - rather than letting them rewrite your own words into a house style that belongs to no one. Writers who want to stay distinguishable in a sea of AI-assisted prose may need to keep more of their own sentences intact. Resources like the AI for Writers tag and the AI Learning Path for Technical Writers offer practical guidance on where that line sits.
Why this matters for writers
The study gives professional writers a concrete reason to reconsider their editing workflow. If you use an AI assistant to rewrite your drafts, your writing is likely becoming less distinctive - and readers can sense it, even if they can't name it. The solution isn't to reject the tools entirely. It's to decide which parts of the writing process benefit from automation and which parts should stay human. Your voice is the one asset that no AI can replicate. This research suggests that using AI to polish everything may quietly erase it.
Your membership also unlocks:
