Cognitive load–driven instruction simplification with generative AI for inclusive manufacturing

Text-based instructions in manufacturing can impose high cognitive demands, particularly for neurodivergent users. This study introduces a methodology for simplifying technical instructions using generative artificial intelligence, guided by cognitive load estimates derived from physiological data. Tasks were represented by features including mean and variable cognitive load, execution efficiency and SMOG readability, and clustered to identify high-demand instances. Instructions from this subset were reformulated using a prompt-based language model applying lexical and syntactic constraints. Readability gains were quantified using the SMOG index, and regression analysis was used to assess the relationship between instruction complexity and cognitive performance. Results indicate reduced linguistic complexity and a non-linear association between readability and cognitive load, suggesting the potential of generative tools for inclusive instruction design. Application 1