Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-31 · high
A Nature Human Behaviour study of 880,000 texts found LLM polishing cuts writing-complexity variance 21 to 50%, erasing cues to gender, age and ideology.
Why now: The study, newly published in Nature Human Behaviour, ties the drop in writing variance to a specific before-and-after point, ChatGPT's release.
LLM polishing cuts variance in how people write by 21 to 50%, per a Nature Human Behaviour study of 880,000 texts. It matters most for fields that score writing style directly. Psychology assessment, mental health screening and recruitment all treat word choice and sentence structure as input to a decision.
The pattern held across three studies and seven datasets. Polishing strips out small cues to gender, age, ideology and moral values. Dominant patterns in the training data get amplified; less common ones get suppressed. Text that's been through a model reads cleaner, and it starts to sound like everyone else's.
It doesn't say how much polishing triggers the effect, or whether a light grammar pass matters as much as a full rewrite. It also doesn't say whether the compression reverses once someone stops using the tool. Those are open questions for anyone relying on writing style as a signal right now.
Each link below shares sources, entities, or timing with this story.
LLM uses OpenAI / Shared entities / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (LLM uses OpenAI); both cover ChatGPT, LLM; overlapping topics (against, chatgpt).
Simon Willison released LLM / Shared entity: ChatGPT / Shared topic / Earlier coverage / Downstream implication
Linked by a graph relationship (Simon Willison released LLM); both cover ChatGPT; overlapping topics (chatgpt, cover, finding).
LLM uses OpenAI / Shared entities / Earlier coverage / Tension
Linked by a graph relationship (LLM uses OpenAI); both cover ChatGPT, LLM; earlier ChatGPT coverage from 2026-07-26.
LLM uses OpenAI / Shared entity: ChatGPT / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (LLM uses OpenAI); both cover ChatGPT; overlapping topics (against, chatgpt).
Simon Willison released LLM / Shared entity: LLM / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; overlapping topics (against, concrete).
LLM uses OpenAI / Shared entity: ChatGPT / Shared topic / Earlier coverage
Linked by a graph relationship (LLM uses OpenAI); both cover ChatGPT; overlapping topics (chatgpt, cover, finding).
LLM uses OpenAI / Shared entities / Earlier coverage
Linked by a graph relationship (LLM uses OpenAI); both cover ChatGPT, LLM; earlier ChatGPT coverage from 2026-08-21.
Linked by a graph relationship (LLM uses OpenAI); both cover ChatGPT, LLM; earlier ChatGPT coverage from 2026-03-21.