Artificial Intelligence in Scientific Publishing: A Narrative Review of Writing, Peer Review, Ethics, Equity, and Governance

Abstract

Artificial intelligence (AI) is increasingly reshaping scientific publishing by supporting manuscript preparation, language editing, editorial screening, and peer review while introducing new ethical, methodological, and governance challenges. This narrative review examines the emerging role of large language models (LLMs), including ChatGPT, across scientific writing and editing, editorial workflows, disciplinary adoption, research integrity, publishing equity, and institutional policy. A structured search of the Scopus database identified 152 records, of which 22 studies met the eligibility criteria and formed the primary synthesis. The reviewed evidence indicates that AI can improve readability, linguistic clarity, translation, and the efficiency of structured editorial tasks; however, these benefits do not consistently translate into greater technical accuracy, analytical depth, or scholarly judgement. Adoption also varies across disciplines and publishing contexts. Major concerns include citation hallucination, factual inaccuracies, authorship ambiguity, bias, confidentiality, inadequate disclosure, and over-reliance on automated outputs. At the same time, AI-assisted language and translation tools may reduce barriers faced by non-native English-speaking and under-resourced researchers, although unequal access to advanced systems may reinforce existing disparities. Current journal and institutional policies increasingly converge on disclosure, prohibition of AI authorship, reference verification, and human accountability, although implementation remains inconsistent. Overall, the evidence supports the use of AI as an assistive rather than autonomous scholarly tool. Responsible integration requires transparent disclosure, rigorous verification, meaningful human oversight, equitable access, and adaptive governance.