Not all information helps equally: context determines what supports effective action. This Situated Action principle might matter for Augmented Reality (AR), as different designs might be used in response to different user states. Accordingly, we investigate how stress and guidance level jointly influence performance and acceptance through a 2 2 experiment ( ) in AR-assisted assembly. Participants underwent stress induction or relaxation before using AR applications with high (step-by-step) or low (level-based) guidance. Our findings challenge simple stress-regulation models: guidance level reconfigured the pathways to outcomes. Under high guidance, stress predicted faster completion but reduced hedonic motivation and acceptance. Under low guidance, performance decoupled from stress, with utilitarian and hedonic motivations driving both efficiency and reuse intention. Structural equation modeling supports these guidance-contingent mechanisms. By introducing a Situated Action perspective, we demonstrate that different AR applications can reshape how tasks are solved, potentially impacting AR application design as well as future context-adaptive and pervasive AR systems. These implications are discussed. Our study offers a theoretical basis for hypothesis development and advances theory on how AR shapes user behavior.