Polished language may preserve meaning while quietly weakening the personal signals carried by words.
LOS ANGELES, UNITED STATES
Artificial intelligence writing assistants can make texts clearer and more grammatically consistent, but they may also reduce the linguistic differences that distinguish one writer from another. A study published in Nature Human Behaviour examined more than 880,000 texts and found that writing styles became less varied as the use of large language models expanded. The research was conducted by Zhivar Sourati and colleagues, principally affiliated with the University of Southern California.
The researchers analysed academic abstracts from arXiv, creative stories posted on Reddit and articles from the US local-news network Patch. In the datasets covering periods before and after ChatGPT’s public release, increased indications of AI involvement were associated with narrower differences in vocabulary, sentence structure and writing complexity. Because this part of the research was observational, it identifies an association rather than proving that AI alone caused every change.
Controlled experiments provided stronger evidence of the homogenising effect. Human-written texts were rewritten by GPT-3.5, Llama 3 and Gemini Pro using instructions such as improving grammar, readability or naturalness. The systems generally preserved the original meaning, but reduced variation in writing complexity by between 21 and 50 percent. The convergence appeared across different models, prompts and text categories.
AI-assisted rewriting also made personal characteristics more difficult for computational classifiers to infer. Predictions involving age, gender, personality, empathy and moral values became less accurate after the texts were processed. The signals were not completely erased, however, and the systems introduced their own directional biases. Rewritten authors were more frequently classified as older, more agreeable and open, but less empathetic and extroverted.
These findings do not establish that software can reliably determine a person’s identity or psychological condition from a single document. The researchers measured changes in statistical linguistic patterns, not definitive diagnoses of individual writers. Their results nevertheless raise concerns for fields that use language to evaluate candidates, personalise services, study cultures or identify possible psychological risks.
The central problem is therefore not simply whether AI produces recognisable prose. It is whether repeated dependence on the same systems gradually replaces diverse human voices with a narrower model of what clear, professional or acceptable writing should resemble. Preserving authorship may require writers to use AI selectively, retain control over revisions and consciously protect their vocabulary, rhythm and cultural perspective.
La verdad es estructura, no ruido. / Truth is structure, not noise.