Scientists Have Been Quietly Using AI to Write Research Papers
A massive analysis of 15 million biomedical abstracts has revealed that AI-generated content has quietly entered academic publishing at unprecedented levels since the launch of ChatGPT.
How Scientists Caught AI Red-Handed
Researchers from the U.S. and Germany took a detective-like approach to uncover AI’s influence on scientific writing. Rather than trying to detect AI-generated text directly, they analyzed word patterns in biomedical abstracts on PubMed from before and after large language models (LLMs) became widely available.
The team based their method on how COVID-19 public health studies tracked excess deaths to assess the pandemic’s impact. Instead of excess deaths, they tracked “excess words” to identify AI’s footprint in academic writing.
Their findings show a clear shift since ChatGPT’s release less than three years ago. Academic papers now contain more “stylistic and flowery” language, a marked change from the previously favored concrete “content words.” Words such as “pivotal,” “grappling,” and “showcasing” have surged in usage. According to the study published in Science Advances, at least 13.5% of papers published in 2024 involved some form of AI assistance.
The shift isn’t limited to word choice. Before 2024, 79.2% of excess words were nouns, but that flipped in 2024: 66% became verbs and 14% adjectives. This reflects a deeper change in how papers are structured.
Why Flowery Language Is Raising Red Flags
This trend raises important questions about academic credibility. When researchers use AI to polish their writing, they’re outsourcing part of their scholarly voice to algorithms trained on existing literature. This could lead to peer reviewers unknowingly approving AI-enhanced papers, creating a feedback loop where AI-generated content becomes the academic norm.
Unlike earlier detection methods that compared human and AI-generated samples—approaches that risk bias by assuming which AI models or prompts scientists use—this study’s word-pattern approach avoids those pitfalls.
Notably, the study found differences in AI usage between research fields, countries, and journals, suggesting some academic communities have adopted AI tools faster than others.
Academic Publishing’s New Reality
This research reveals a fundamental shift beneath the surface of academic publishing. AI isn't just changing writing style; it’s altering what gets written and how scientific knowledge circulates.
The 13.5% figure likely underestimates the true extent of AI use. As AI tools improve and become harder to detect, this share could rise substantially.
The Disclosure Debate Ahead
The presence of AI in academic writing is now undeniable. Rather than resisting this change, the scientific community must develop clear guidelines on AI use in research papers.
Some journals have already updated policies requiring authors to disclose AI assistance, while others invest in detection technologies. The key question remains: where should the line be drawn between acceptable AI help and problematic reliance?
Evidence suggests we have moved beyond casual AI use. AI-assisted writing is becoming the new standard, not an exception.
Your membership also unlocks:
