A recent analysis reveals that around 14% of biomedical research abstracts published in 2024 were likely generated with the aid of artificial intelligence (AI), particularly large language models (LLMs).

This study, which examined over 1.5 million abstracts indexed in PubMed, identified more than 200,000 that exhibited language characteristic of AI-generated text.

Conducted by researchers from the U.S. and Germany, the study analyzed over 15 million biomedical abstracts to assess the prevalence of AI-generated content in academic writing.

In 2024, researchers identified 454 'excess words'—stylistically oriented terms rather than content-driven ones—indicating a significant shift in scientific vocabulary influenced by LLMs.

Common stylistic terms that have gained popularity include 'delves', 'showcasing', 'underscores', 'potential', 'findings', and 'critical', reflecting a change in writing style.

The findings also showed a notable shift from content words, primarily nouns, to more stylistic language, with 66% of excess word choices in 2024 being verbs and 14% adjectives.

The researchers employed an innovative method by identifying these 'excess words', which have become more prevalent since the introduction of ChatGPT in late 2022.

To mitigate biases, the study utilized a before-and-after analysis of word usage patterns surrounding ChatGPT's release, akin to methodologies used in public health research.

The impact of AI on scientific writing in biomedical research appears unprecedented, surpassing influences from major global events like the COVID-19 pandemic.

The proportion of abstracts utilizing AI varied by discipline, with some fields reporting up to 40% usage, particularly in computational areas where researchers are more familiar with AI.

Since ChatGPT's launch, there has been a significant rise in AI-generated content online, raising concerns about the accuracy and integrity of research publications.

However, quantifying the impact of LLMs remains challenging due to a lack of transparency from users regarding their use of these AI tools in writing.