AI Revolution and Scientific Research Involving LLMs and Gen-AI: Preliminary PICO-AI, SPIDER-AI, and PRISMA-AI Conceptual Extensions for Evidence Synthesis Frameworks

Background: Systematic and scoping reviews rely on established frameworks such as PICO (Population, Intervention, Comparator, Outcome), SPIDER (Sample, Phenomenon of Interest, Design, Evaluation, Research type), and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) to structure research questions, guide evidence synthesis, and ensure transparent reporting. These frameworks were developed for linear, human-driven workflows involving manual searching, screening, and data extraction. The emergence of large language models (LLMs) and Generative AI (Gen-AI) introduces new capabilities for semi-automated literature discovery, screening, data extraction, and thematic synthesis. However, the methodological implications of AI-assisted evidence synthesis remain underexplored. Methods: This manuscript employed a scoping review–informed methodological analysis of empirical and methodological literature located through structured searches of PubMed and Google Scholar. Preliminary conceptual extensions to PICO, SPIDER, and PRISMA were developed, retrospectively applied to published studies, and independently evaluated by five researchers for applicability, relevance, and clarity. Results: LLMs enable dynamic and iterative processes across evidence synthesis stages, including automated question formulation, large-scale qualitative analysis, and hybrid human–AI screening and extraction. These developments challenge the linear assumptions of PICO, SPIDER, and PRISMA. In response, PICO-AI, SPIDER-AI, and PRISMA-AI incorporate dimensions including prompt transparency, model specification, human-in-the-loop validation, Human–AI workflow classification, data governance, and AI auditability. Independent evaluation showed 80% item agreement, with overall mean relevance and clarity ratings of 3.2/4. Conclusions: LLMs and Gen-AI do not replace systematic review frameworks; rather, they extend them by introducing adaptive, data-driven workflows. Integrating AI into evidence synthesis requires revised methodological guidance to ensure transparency, reproducibility, and rigour. The proposed preliminary extensions provide a model for incorporating machine intelligence into systematic and scoping review practices.