Student researchers increasingly rely on large language model (LLM)–based AI tools, such as ChatGPT, to support academic reading. Some scholars warn that students may offload core cognitive tasks to LLMs, thereby losing opportunities to practice ess...
Student researchers increasingly rely on large language model (LLM)–based AI tools, such as ChatGPT, to support academic reading. Some scholars warn that students may offload core cognitive tasks to LLMs, thereby losing opportunities to practice essential competencies such as critical evaluation of papers and argument construction. However, the effects of LLM assistance on critical thinking during academic paper reading remain unclear, leaving educators with limited guidance on how to integrate these tools appropriately. In this paper, I investigate how LLM assistance influences critical thinking in academic paper reading tasks. I hypothesize that the impact of LLM assistance on student researchers’ critical thinking is moderated by their level of research experience. I conducted a within-subjects experiment with 13 student HCI researchers, who read academic papers with and without LLM assistance and wrote critical reviews. I observe divergent effects: LLM assistance improves critical thinking among experienced student researchers but tends to undermine it among novices, with substantial individual variability. Based on these findings, I discuss design implications for preserving human agency while delegating cognitive labor to LLMs in academic work.