AI Insight
This study examined how 1,760 terminology professionals using large language models structure their prompts and how this relates to their cognitive workload and confidence in AI interaction. Results showed that practitioners who use more structured prompting strategies experience lower cognitive load, with this relationship mediated by their self-efficacy in working with AI. These patterns remained consistent across all experience levels, from novices to experts, suggesting universal cognitive mechanisms in human-AI collaboration for knowledge work.
Why it matters
The findings provide empirical evidence that training workers to use structured prompting techniques could reduce mental strain and improve confidence when working with generative AI tools in terminology and similar knowledge-intensive fields. This has practical implications for designing professional training programs and workplace interventions as organizations increasingly adopt AI systems.
Understand the Science
IntroductionThe widespread application of generative artificial intelligence (AI) in knowledge-intensive domains requires continuous adjustment of human cognitive patterns. In terminology work, aligning strict semantic standards against the non-deterministic output of large language models (LLMs) is closely intertwined with workers’ subjective cognitive load. This study examines the structural associations among prompt strategy preference, AI interaction self-efficacy, and subjective cognitive load, while testing the potential moderating role of professional experience.MethodsBased on cross-sectional survey data from 1,760 practitioners using LLMs for terminology tasks, structural equation modeling (SEM) and latent profile analysis (LPA) were employed. SEM evaluated the linear structural associations among prompt strategy preference, self-efficacy, and cognitive load, while LPA identified behavioral user profiles. Multi-group equivalence testing was further conducted across three career stages (novices, intermediates, and experts).ResultsPrompt strategy preference is significantly and negatively associated with subjective cognitive load, with AI interaction self-efficacy serving as a significant mediator in this structural relationship. LPA identified three distinct user profiles (low, medium, and high strategy use), with the high strategy group exhibiting the lowest cognitive load and highest self-efficacy. Multi-group equivalence tests indicated no statistically significant moderation across experience cohorts, demonstrating structural invariance of the core model across career stages.DiscussionFrom the perspective of cognitive ergonomics, upfront prompt structuring is structurally linked to downstream cognitive offloading and psychological buffering. These findings provide empirical evidence for individual cognitive resilience during human-AI collaboration in the generative AI era.