Under pressure : the influence of psychological and cognitive factors in deepfake detection under time pressure

Abstract

Advances in AI have produced increasingly realistic deepfakes, which are exploited in social engineering attacks. Studies show human ability to detect these fake videos is generally poor and varies widely (accuracy 23–87%). Time pressure further undermines performance by depleting cognitive resources and encouraging heuristic processing. This thesis examines psychological and cognitive factors affecting deepfake detection under time constraints. Specifically, it investigates how Big Five personality traits, sense of coherence (stress resilience), cognitive workload, and prior deepfake detection experience influence accuracy in a time-pressured deepfake video evaluation task. A quantitative explanatory design with 89 participants was used. Participants completed measures of personality (BFI-10) and sense of coherence (SOC-13), reported cognitive workload (NASA-TLX), and then performed a timed deepfake detection task adapted from Köbis et al. (2021). Each participant viewed ten short videos (five real, five deepfake), only once, under time pressure, and judged each as real or fake. Multiple regression and fuzzy-set qualitative comparative analysis (fsQCA) assessed how these factors predicted detection accuracy. Regression analysis revealed that only prior detection experience significantly predicted accuracy (β ≈ .04, p = .001), whereas no personality trait, sense of coherence score, or workload rating had a significant effect. FsQCA identified multiple distinct configurations of conditions associated with both high and low detection accuracy, underscoring the principle of equifinality in performance outcomes. Notably, the personality traits openness to experience and neuroticism emerged as important contributors within several configurations. Notably, participants were far more accurate in identifying real videos (≈75% correct) than deepfakes (≈48% correct), reflecting a truth-bias under time pressure. These findings suggest that personality, cognitive workload, and general stress resilience play a more nuanced role in deepfake detection under pressure, whereas hands-on experience is critical. Accordingly, organizations should implement targeted training interventions tailored to employees to improve their ability to detect deepfakes under time constraints. Such focused training can complement technical defences and strengthen organizational resilience to deepfake-based social engineering. Recommendations: Develop and deploy tailored training programs in organizational settings that build deepfake detection skills under pressure, adapting content to employee experience to enhance real-world detection accuracy.