Delving Into PubMed Records: How AI-Influenced Vocabulary has Transformed Medical Writing since ChatGPT

Area of Science:

• Medical Literature Analysis
• Natural Language Processing
• Academic Writing Trends

Background:

• Large language models (LLMs), including ChatGPT, are increasingly utilized in academic writing.
• Concerns exist regarding the potential homogenization of scientific language due to AI tool usage.
• This study investigates the prevalence of specific LLM-associated terms in medical research.

Purpose of the Study:

• To determine if words and phrases frequently used by LLMs have increased in medical literature.
• To compare the usage trends of potentially AI-influenced terms with common academic expressions.
• To understand the timeline of these vocabulary shifts in relation to LLM development.

Main Methods:

• A systematic literature review identified 135 potentially AI-influenced terms from 15 studies.
• 84 common academic phrases were selected as control terms.
• PubMed records from 2000-2024 were analyzed, with term frequencies normalized using modified Z-scores.

Main Results:

• 103 of 135 potentially AI-influenced terms showed significant increases (modified Z-score ≥3.5) by 2024.
• Terms like "delve," "underscore," "primarily," "meticulous," and "boast" exhibited notable growth.
• Usage of these terms began rising in 2020, accelerating significantly in 2023-2024, preceding ChatGPT's release.

Conclusions:

• The increased use of certain terms in medical literature appears to be an amplification of pre-existing trends by LLMs.
• Understanding AI-driven vocabulary shifts is crucial for medical educators and researchers.
• Promoting diverse vocabulary and careful editing of AI-assisted drafts is recommended to maintain scientific writing quality.