Table of Contents
AI may be changing nature of writing, narrowing public discourse
A new study by researchers from institutions, including the University of California, Berkeley, University of Washington and Google DeepMind suggests artificial intelligence (AI) tools are not just improving writing, but actively altering its meaning.
With more than a billion people using AI tools weekly – much of it for writing – the influence is already widespread. What appears to be a productivity tool is beginning to shape how ideas are expressed and possibly how they are formed.
Editing that changes meaning
At first glance, AI writing assistants function like advanced spellcheck tools. But the study finds their impact goes far beyond grammar.
In controlled experiments from the study How LLMs Distort Our Written Language, researchers compared essays written by humans with those written or edited using AI. Even when models were told to make only minimal edits, they still introduced noticeable shifts in meaning.
These were not random changes. Across multiple models and prompts, AI revisions consistently pushed writing in similar directions, unlike human edits, which tended to be smaller and more varied. The result is not just cleaner writing but different writing.
Drift toward neutrality
One of the clearest patterns is a shift in opinion. In a user study, participants who relied heavily on these tools were far more likely to produce neutral essays, rather than taking a clear stance.
Instead of arguing for or against a topic, their writing drifted toward balance and ambiguity. This suggests that AI tools may be smoothing out strong viewpoints, even when users intend otherwise.
At the same time, participants reported that their work felt less like their own. Heavy AI users were more likely to say their writing lacked creativity and did not reflect their personal voice, even if they were still satisfied with the final result.
Rise of shared voice
Beyond meaning, the study points to a broader stylistic shift. Large language models tend to produce a shared voice, where different users’ writing begins to converge.
Essays generated with heavy AI use showed similar structure, tone and vocabulary, while human-written texts remained more diverse. This points to a gradual flattening of individual expression.
The shift also appears in grammar and tone. AI-assisted writing uses fewer personal pronouns and more nouns and adjectives, moving toward a more formal and impersonal style. Writing becomes more polished, but also more detached from individual identity.
More persuasive, less personal
The models also reshape how arguments are built. AI-generated text relies more on structured reasoning, statistics and generalised claims rather than personal experience.
At the same time, it increases the use of emotional language, particularly words associated with positivity and trust. This mix of analytical framing and emotional reinforcement can make writing more persuasive.
However, it is not always more faithful to the original intent. The concern is not obvious errors, but subtle reframing that can be difficult to detect.
Shifting decisions, not just words
The implications extend beyond everyday writing. Researchers analysed peer reviews from a major AI conference and found that a significant portion were generated or heavily edited by AI.
These reviews differed from those written by humans. They tended to assign higher scores and placed less emphasis on clarity and significance, focusing instead on technical aspects such as scalability and reproducibility.
This suggests that AI is not only influencing how people write, but also how they evaluate work and make decisions.
Slow narrowing of expression
The broader concern is cumulative. If AI tools consistently nudge writing toward neutrality, formality and a shared structure, they may gradually narrow the range of expression in public discourse.
Over time, this could affect journalism, academic work and political communication. The risk is not a sudden loss of originality, but a slow drift toward uniformity.
Control or convenience
The study does not argue against using AI. Instead, it highlights a gap between perception and reality. Users tend to see these tools as passive assistants, but the evidence suggests they play a more active role in shaping content.
The more heavily they are used, the stronger the effect becomes. Those who use AI sparingly, for example as a research or brainstorming tool, show far less distortion in their writing.
As AI becomes embedded in everyday workflows, the question is no longer whether it changes writing. It is how much influence users are willing to accept over their own voice and meaning.
READ MORE:
Hallucinations in AI: how to be certain