Information overload, cognitive fatigue, and postgraduates’ intention to use generative artificial intelligence for English academic writing: the role of lower-order cognitive offloading

Background Generative artificial intelligence (AI) returns large amounts of information that postgraduates must judge and integrate when they write in English. The stressor-strain-outcome framework suggests that such overload strains cognitive resources and weakens plans for further use. This study tested whether cognitive fatigue links perceived information overload with later behavioral intention to use AI for English academic writing. It also tested whether lower-order cognitive offloading, such as having AI summarize papers, conditions the overload-fatigue association. Methods In total, 475 Chinese master’s and doctoral students who used AI for English academic writing completed three surveys at two-week intervals. Overload and offloading were measured at Time 1, fatigue at Time 2, and intention at Time 3. A moderated mediation model adjusted for baseline fatigue, baseline intention, and AI-use frequency. Results Higher perceived information overload accompanied higher cognitive fatigue 2 weeks later and weaker behavioral intention 4 weeks later, net of baseline levels. Cognitive fatigue was associated with weaker later intention and carried part of the overload-intention association. The overload-fatigue association was weaker at higher levels of lower-order cognitive offloading. Among the most frequent offloaders it was not statistically significant, which does not indicate its absence. The indirect association through fatigue was likewise weaker at higher offloading. These patterns held with additional covariates, a competing interaction with AI-use frequency, and shortened scales. Conclusion In this sample, cognitive fatigue accompanied the prospective association between overload and weaker plans to use AI. This pathway was weaker among postgraduates who delegated foundational tasks more often. Writing programs can teach students which foundational tasks to delegate, and tool developers can limit output overload.