ChatGPT’s Linguistic Imprint: How AI is Reshaping the Language of Academic Writing

A comprehensive new study by the University of Wolverhampton’s Statistical Cybermetrics and Research Evaluation Group and the University of Sheffield’s School of Information, Journalism and Communication reveals just how profoundly large language models (LLMs) like ChatGPT are reshaping academic writing. The authors examined six major databases, Scopus, Web of Science, PubMed, Dimensions, OpenAlex, and PubMed Central (PMC), covering more than 2.4 million full-text open-access biomedical papers between 2021 and mid-2025. Their focus was on a dozen words and phrases frequently associated with AI-generated text, and the results show unmistakable stylistic fingerprints of machine influence. Words such as delve, underscore, and intricate have exploded in popularity since ChatGPT’s launch in late 2022, marking a linguistic revolution in the halls of scholarship.

A Surge of AI-Linked Vocabulary

The data tells a striking story. In Scopus and Web of Science, delve rose by more than 1,500 percent between 2022 and 2024, while underscore increased by over 1,000 percent and intricate by 700 percent. By 2025, nearly 30 percent of PMC papers used underscore, compared with just 3 percent in 2022. Review articles leaned especially heavily on the term, with almost half including it in their pages. Other words like pivotal, meticulous, and showcase have followed the same trajectory. This shift goes far beyond occasional appearances in abstracts: these words are now deeply embedded in full texts, pointing to a broad adoption of AI-shaped writing patterns.

The growth is uneven, however. STEM disciplines are adopting this style most aggressively, with increases above 3,000 percent in computer science, engineering, physics, and mathematics. Social sciences, business, and arts and humanities show slower uptake, though some already used terms like delve more frequently before 2022. Environmental science offers a vivid example: the appearance of underscore grew fifteen-fold in just two years, showing how quickly LLM-associated vocabulary can infiltrate entire fields.

From Rare Occurrences to Repeated Use

What makes the transformation even more compelling is not just the rising presence of these words but their repeated use within the same papers. In 2022, it was unusual to see any LLM-style word more than once or twice in a single article. By 2025, the situation had changed dramatically. The number of papers using underscore six or more times increased by over 10,000 percent. Intricate and meticulous were not far behind, with jumps of 5,400 percent and 2,800 percent. These terms are no longer stylistic flourishes, they have become structural features of scholarly prose.

Patterns of co-occurrence tell the same story. In 2024, 59 percent of PMC papers that used delve also used underscore, compared with just 1.3 percent in 2022. Once, negligible correlations have become strong and consistent. Underscore is now tightly linked with pivotal, intricate, and nuance, forming a bundle of AI-shaped expressions that increasingly appear together. In short, when one of these words enters a paper, others are likely to follow, revealing a stylistic package reminiscent of ChatGPT’s default phrasing.

Outpacing Traditional Academic Terms

The researchers compared these AI-associated words with their more traditional academic counterparts and found striking contrasts. Between 2022 and 2024, delve shot up 1,360 percent, while investigate grew by just 9 percent. Underscore soared by more than 1,000 percent, while highlight managed only 85 percent. Intricate and meticulous outpaced complex and precise by large margins. Such differences suggest that the rise of AI-linked vocabulary cannot be explained by broader linguistic drift alone; it points to a stylistic revolution seeded by LLM outputs.

Yet the picture is not entirely celebratory. The study shows that by 2024, 15 percent of retracted papers contained at least one LLM-associated term, compared with 8 percent of published papers. While this does not prove causation, it raises questions about whether overreliance on AI-generated text correlates with weaker scholarship or questionable practices. The authors stress that many legitimate uses, such as proofreading and translation, especially by non-native English speakers, may also drive the surge and should not be conflated with misconduct.

A New Linguistic Era in Academia

Despite limitations, the focus on twelve English terms, the emphasis on biomedical open access literature, and the exclusion of journals like The Lancet or The New England Journal of Medicine, the scale of the data makes the evidence compelling. Millions of articles across six databases show the same trajectory: LLMs are changing how science sounds. The implications are double-edged. On the one hand, these tools are lowering barriers to publication for scholars worldwide, giving non-native English speakers a stylistic boost and accelerating the flow of ideas. On the other hand, the flood of publishable material could strain journals, raise rejection rates, or force new publishing models.

Academic writing has entered a new linguistic era, with artificial intelligence increasingly dictating the rhythm and vocabulary of scholarship. The language of research is becoming subtly but unmistakably AI-inflected. Whether this development ultimately enriches science or dilutes it will depend on how researchers, editors, and institutions navigate the balance between accessibility, originality, and rigor. What is clear, however, is that the invisible hand of LLMs is already at work, quietly reshaping the texture of academic prose in ways that will continue to ripple through universities, journals, and global knowledge production in the years ahead.