Generative Artificial Intelligence in Oncology Publishing: Vocabulary Adoption Without Gains in Publication Volume, 2018–2025, a Two-Stream Bibliometric Analysis with Implications for Hospital Library Practice

Since the public release of ChatGPT in November 2022, hospital librarians have faced increasing questions about whether artificial intelligence (AI) tools are changing the biomedical literature they navigate for patrons. This bibliometric study examined oncology publishing in PubMed from 2018 to 2025 using two analytical approaches. In total, 1,200 randomly sampled abstracts and 1,131,539 oncology PubMed records were analyzed. The vocabulary analysis found that large language model (LLM)-associated vocabulary in abstracts from 10 major oncology journals nearly tripled between 2020 and 2025, confirming progressive diffusion of AI writing tools. The publication volume analysis found that oncology original research articles declined below their pre-AI projected trajectory by 7.2%, 12.9%, and 20.1% in 2023, 2024, and 2025. Non-original content declined proportionally more, with its share of total output falling from 30.6% to 29.0%. These are associational findings; causation cannot be established from observational bibliometric data. These findings have direct implications for hospital librarians’ PubMed search strategies, evidence appraisal, patron counseling, and institutional research support.