AI-Assisted Peer Review
A framework for more efficient and reliable scientific evaluation
Spiros Kitsinelis
Association of Greek Chemists
eex@eex.gr
DOI: 10.62579/JAGC0021
Abstract
Peer review is central to scientific quality assurance, yet it faces persistent challenges, including reviewer workload, delays, variability in assessment and potential bias. Artificial intelligence (AI), particularly large language models, may support selected aspects of manuscript evaluation, but its reliability, limitations and appropriate role in editorial decision-making require careful consideration. This Perspectives article examines the potential of AI-assisted peer review, drawing on preliminary observations from an exploratory exercise in which two independent AI systems evaluated a number of scientific manuscripts. Both systems identified potential weaknesses in methodological reporting, data presentation and statistical analysis, while also offering differing emphases on clarity, interpretation and context. Because the exercise was not benchmarked against expert human reviewers and did not use a formal quantitative evaluation protocol, these observations are illustrative rather than evidence of validated reviewing performance. Building on these observations, we propose a twelve-criterion framework for structured manuscript assessment and a human-supervised workflow in which AI supports screening and the identification of potential problems, while expert reviewers and editors retain responsibility for evaluating scientific significance, interpreting disputed findings and making publication decisions. The framework emphasizes independent verification, confidentiality, transparency and accountability. Blinded, controlled studies are needed to establish the accuracy, consistency, fairness and practical value of AI-assisted review. Responsible integration should therefore be guided by empirical validation rather than assumptions that AI is inherently objective or that faster generation of a review necessarily improves the publication process.
Keywords: artificial intelligence; peer review; scientific publishing; research integrity; reproducibility; editorial governance
Full article: