USE OF AI IN WRITING RESEARCH PAPERS: A PRISMA‑GUIDED SYSTEMATIC LITERATURE REVIEW AND GOVERNANCE FRAMEWORK

Abstract

This systematic review examines the utilization of generative artificial intelligence in the composition of research papers and explores the responsible governance of such practices. Adhering to the PRISMA 2020 guidelines, we conducted searches across multidisciplinary databases and publisher policy portals from 2018 to August 2025, screened records, assessed eligibility, and qualitatively synthesized the findings. Thirty-six sources met the inclusion criteria, encompassing randomized and field studies on writing productivity and quality, evaluations of citation reliability and detector bias, and formal policies from major editorial bodies and publishers. The studies indicate that AI assistance enhances drafting speed and perceived clarity, with the most significant improvements observed among writers with lower initial proficiency and in micro-revision tasks. Risks are primarily associated with fabricated or mismatched references, subtle factual inaccuracies, loss of disciplinary voice, confidentiality breaches when using public tools, and false positives from AI-text detectors affecting non-native writers. Policies converge on three norms: prohibition of AI authorship, mandatory disclosure of substantive use, and full human accountability for content and citations. We propose HILSA 2.0, a human-in-the-loop workflow incorporating evidence-verification gates, disclosure ledgers, and role-specific responsibilities for authors, supervisors, and journals. The future research agenda prioritizes randomized trials on scholarly outcomes, equity audits of detector policies, provenance methods for AI-assisted text, and longitudinal effects on skill development. Within the scope of the abstract, nuances and boundary conditions are explicitly delineated to facilitate replication. For clarity, we define generative AI as systems that produce text conditioned on prompts using large language models.