A new study of nearly 5,000 student reports shows that writing has grown more polished, formal and upbeat since ChatGPT’s launch, even as grades remain unchanged. The findings raise new questions about voice, critical thinking and how universities should teach writing in an AI-rich world.
Student writing has become more polished, formal and upbeat since the launch of ChatGPT — but grades have barely budged, according to a new study led by the University of Warwick.
Analyzing almost 5,000 student-authored reports over a decade, the researchers found that the language students use has shifted noticeably since late 2022, when generative artificial intelligence tools such as ChatGPT became widely available. Their work, published in the journal Computers and Education: Artificial Intelligence, suggests that AI is subtly reshaping how students write, even if it has not yet transformed what they are able to argue or explain.
The team examined 4,820 empirical reports containing about 17 million words, submitted by undergraduates over a 10-year span. Rather than trying to catch individual students using AI, the researchers looked at broad, cohort-level patterns in style, tone and vocabulary before and after ChatGPT’s arrival.
They found that, starting in 2022, student writing became more positive in sentiment, more formal in tone and more varied in vocabulary. Those changes appeared across reports regardless of topic, pointing to a general shift in how students are putting words on the page.
First author Matthew Mak, an assistant professor of psychology at Warwick, said the tone of student work now closely resembles the default style of popular AI tools.
“The tone of students’ writing appears more positive, in line with ChatGPT’s output, which is not inherently a good or bad thing, but it does raise concerns about the possibility of AI tools homogenising students’ voices,” Mak said in a news release.
Many generative AI systems are designed to sound polite, constructive and upbeat, in part to avoid offensive or harmful content. That default positivity may be rubbing off on students who use these tools to brainstorm, rephrase or polish their assignments.
Mak and his colleagues note that psychological research has linked positive mood to lower levels of critical scrutiny. If students are constantly exposed to AI-generated text with a relentlessly optimistic tone, the researchers argue, universities need to understand how that might shape students’ long-term critical thinking and academic voice.
The study also detected a sharp rise in formality and vocabulary range after ChatGPT’s launch. Those are the kinds of stylistic improvements that typically emerge only after many years of writing practice, which makes it unlikely they reflect a sudden leap in students’ underlying skills.
In other words, the essays may look more sophisticated on the surface, but that does not necessarily mean students have become better writers in the deeper sense of structuring arguments, weighing evidence or interpreting data.
To probe the influence of AI more directly, the researchers took reports written before 2022 and asked ChatGPT to rewrite them. The AI-generated rewrites showed the same kinds of shifts in tone and style that appeared in the real student work submitted after ChatGPT became available. That parallel pattern strengthens the case that generative AI is a key driver of the changes the team observed.
The researchers also tracked the use of words that are often associated with AI-generated prose. Some of those terms spiked in popularity through 2024 and then dropped sharply in 2025. That pattern suggests students may have become more cautious about sounding like a chatbot, perhaps in response to growing awareness of AI-detection tools or evolving university policies.
Despite all these stylistic shifts, one thing did not change: grades.
Across the 10-year dataset, there were no corresponding jumps in marks or in examiner feedback that would indicate a broad improvement in the quality of students’ reasoning or analysis. The authors argue that this stability is a sign that core academic skills still matter most.
In a detailed summary of the results, the research team wrote that “despite the stylistic shifts, the grades and feedback received by student remained relatively stable. This suggests that the stylistic shifts do not necessarily translate into higher scores and that our markers may prioritise factors such as analytical depth and data interpretation over surface-level language.”
That finding may come as a relief to educators worried that AI would instantly upend assessment. At the same time, it underscores a growing gap between how polished student work looks and what it actually demonstrates about learning.
The study arrives as generative AI use on campus continues to surge. A recent sector-wide survey cited by the Warwick team found that up to 88% of students report using ChatGPT for assessments. Those uses range from idea generation and outlining to full drafts and editing.
For universities, the Warwick researchers say, the new data is both a warning and an opportunity.
Their analysis highlights that AI is already changing the texture of student writing, even when it does not dramatically alter grades. That reality, they argue, should push institutions to rethink how they design assignments, teach writing and talk with students about responsible AI use.
In their discussion of the implications, the authors emphasized that “core academic skills — such as critical reasoning, interpretation, and argumentation — remain central to assessment” and that these have “at least, not yet been overshadowed” by stylistic changes. That leaves room for educators to double down on teaching those deeper skills while also helping students develop authentic voices in an AI-rich environment.
The findings also suggest that simple surface-level checks — such as looking for certain buzzwords or a uniformly upbeat tone — will not be enough to understand whether and how students are using AI. Instead, faculty may need to focus more on process: how students plan, draft and revise, and how they can articulate their own thinking.
The authors argue that institutions now have a chance to get ahead of the curve, using evidence like this to guide policy rather than reacting piecemeal to each new tool. That could mean clearer guidance on when AI is acceptable, more transparent conversations about academic integrity and new forms of assessment that emphasize oral defense, in-class writing or iterative feedback.
As generative AI becomes a routine part of knowledge work in many fields, the question for higher education is no longer whether students will use these tools, but how. The Warwick study suggests that, so far, AI is changing the look and feel of student writing more than its substance — and that what universities do next will help determine whether those changes ultimately strengthen or weaken students’ ability to think and communicate for themselves.
Source: University of Warwick