Responsible AI in the research lifecycle: a PRISMA-ScR scoping review (2024–2025)

Abstract

Generative artificial intelligence is reshaping scholarly work by expanding productivity while demanding rigorous safeguards for research integrity. This scoping review, aligned with PRISMA ScR, maps responsible integration of AI across the research lifecycle, including search, writing, peer review, reporting, and editorial governance, and synthesizes practices for universities and journals. A structured Scopus search limited to English open access and final status during 2024 and 2025 identified seventynine records. After screening, fifteen documents were included, spanning editorial policies and statements, empirical and meta-research, reporting guidance, and methodological proposals. Analysis revealed convergence on core principles. Artificial intelligence cannot be credited as an author. Disclosure must be explicit and granular, specifying tool, model version, intended purpose, and prompts when relevant. Given known limitations and biases, AI detectors should not constitute sole evidence in editorial decisions. Policies are evolving toward tiered use with human accountability and confidentiality safeguards. In peer review, uploading unpublished manuscripts to public services is discouraged, and permitted tasks for language models are narrowly defined under supervision. Reporting checklists clearly require identification of model and version, seeds, prompts, and evaluation criteria. Version controlled workflows enable audit trails of AI and human contributions. Methodological risks, including instability from prompt repetition, underscore the need for reproducibility and robust traceability. We propose a minimal architecture for responsible adoption that implements disclosure matrices, human accountability, a limited role for detectors, supervised peer review protocols, reporting checklists, and transparent change logs. This architecture strengthens integrity, reproducibility, and trust across scholarly communication.