Use of generative artificial intelligence in medical libraries: A scoping review

Abstrak

Purpose. This study aims to map how health and medical libraries have used GenAI to support library services. Methodology. Scoping review following PRISMA-ScR. Searches were conducted between 2020 and 2025 in PubMed, Embase, CINAHL, LISTA, Academic Search Premier, Ovid, Scopus, Web of Science, ProQuest Dissertations & Theses, and Google Scholar. Of 250 records, 106 remained after deduplication to screening in Rayyan, blinded, by two reviewers. Results and discussion. We included 11 studies: 7 primary and 4 secondary/methods. Among primaries, 5 directly used GPT-family models (GPT-4/ChatGPT/GPT-4o or Perplexity layered on GPT-4) to: answer clinical questions, support search-strategy design, trial library tasks, and integrate source-linked answers in a national portal. One survey assessed intention to adopt generative chatbots; one case described a rule-based chatbot. GenAI accelerates workflows and extends services but requires human verification, transparency about limitations, privacy policies, and staff training (including prompt engineering). LLM-generated search strategies showed variability and underperformed human designs. Conclusions. In its current state, GenAI works best as a copilot useful for drafting, triage, and initial sense-making, while librarians remain essential for high-precision tasks (performance), quality control (reliability), and governance.