Research trends and comparisons of major generative artificial intelligence platforms for systematic literature reviews

Abstract

A variety of SLR-AIs have emerged since ChatGPT, accelerating the processes of literature search and summarization. Bibliographic database？based SLR-AIs, in particular, are effective as starting points at the outset of research. They improve researchers’ understanding of new concepts, enable the collection of reliable articles, and generate summaries by automating keyword selection. However, they may produce incomplete answers in emerging fields. Nonbibliographic databasebased SLR-AIs, on the other hand, often overlook key references and important findings. More SLR-AIs offer natural language summaries, visualized results, and suggested followup questions. The ongoing technological race in the SLR-AI landscape, marked by advances in LLM reasoning, agent technology, and RAG, remains fierce. Current SLR-AI technology cannot fully replace researchers but rather serves to augment human expertise. Human researchers provide critical analysis, uniquely identify trends, and offer perspectives on future research directions that go beyond mere summarization. Exclusive reliance on SLR-AIs does not foster the growth necessary to become an independent researcher. Based on my experience, it remains difficult to precisely determine the reliability of SLR-AIs. The future of SLR-AI should be grounded in trustworthy and efficient content, aided by responsible AI practices. This study suggests that careful selection of SLR-AIs can optimize the efficiency of literature reviews. Solutions such as Felo Agent, in conjunction with adherence to the four implications discussed above, represent a prudent choice. However, excessive reliance on SLR-AIs that draw from public sources poses immediate research ethics risks, making final, comprehensive human review essential. As more articles are published utilizing SLR-AI, increasing effort will be required to verify their authenticity, which risks becoming a societal burden.