Skip to main content

AI Now Writes 70% of Biomedical Papers: Academia Pushes for Rules, Not Bans

It's no secret that artificial intelligence has crept into nearly every corner of our digital lives, but its footprint in academic publishing is now impossible to ignore. A recent study, which combed through over 1.19 million English-language biomedical papers on PubMed Central, found that by 2025, roughly 77% of these papers showed clear signs of AI-assisted writing or editing. That's a staggering jump from 19% in 2023 and 52% in 2024. In other words, AI has become the silent co-author of the vast majority of biomedical research.

The AI Footprint Varies by Section and Region

Dig a little deeper, and the pattern becomes even more intriguing. The discussion section—where researchers interpret their findings and speculate on implications—shows the highest AI involvement, with about 78% of papers bearing AI's fingerprints. Abstracts follow at 68%, introductions at 63%, results at 58%, and methods at 54%. It makes sense: these are the sections where clear, persuasive prose matters most, and AI excels at polishing language.

Geography also plays a role. Researchers from non-English-speaking countries like South Korea and China are turning to AI at rates exceeding 80%, far outpacing their counterparts in the UK and the US. The reason? For many, AI is a lifeline for improving grammar and expression, leveling the playing field in a world where English dominates scientific communication.

The Double-Edged Sword of AI Assistance

But this convenience comes with risks. Large language models are notorious for generating plausible-sounding but entirely fabricated facts and references. They can present errors with unwavering confidence, which is a dangerous trait in a field where accuracy is paramount. Even more troubling, some authors fail to disclose their use of AI, quietly eroding the trust that underpins scientific credibility.

Regulation Over Prohibition

So, what's the solution? The researchers behind this study argue that banning AI outright is neither practical nor desirable. Instead, they advocate for establishing clear guidelines that govern its use. These would include strict requirements for fact-checking, verifying the authenticity of references, and mandatory disclosure of AI involvement. When AI assistance has become the norm, defining reasonable boundaries is far more critical than simply pulling the plug.

Key Points

  • By 2025, 77% of English biomedical papers show signs of AI assistance, up from 19% in 2023.
  • AI usage is highest in discussion sections (78%) and lowest in methods (54%).
  • Non-native English speakers, particularly in South Korea and China, use AI at rates above 80%.
  • Risks include fabricated facts, fake references, and undisclosed AI use.
  • Researchers recommend guidelines for fact-checking, reference verification, and AI disclosure, rather than an outright ban.