Trusting the black box: Adapting a multidimensional measure of trust in generative AI

As Generative AI (GenAI) becomes integrated into routine workflows, the critical question shifts from adoption to reliance. However, research is hindered by a lack of an adapted measurement tools, as studies often depend on unverified adaptations of traditional automation scales. This study addresses this gap by developing and validating a multidimensional trust scale explicitly tailored for GenAI. We adapted Körber’s Trust in Automation (TiA) questionnaire to the GenAI context and administered it to a large adult sample (N = 2,169). We first examined whether the original TiA structure transferred to the GenAI context. Because it did not, we used exploratory factor analysis to identify the dimensional structure that emerged in this sample. The analysis revealed that the original factor structure did not replicate. Instead, items consolidated into three distinct dimensions: (1) competence and reliability, (2) familiarity, and (3) caution and uncertainty awareness. A key finding was that negatively worded items formed a coherent, verification-oriented component rather than merely reflecting distrust, supported by higher reliability when items were retained in their original direction. Moreover, distinct patterns of association with GenAI experience further validated the structure: while experience predicted competence and familiarity, it showed weaker associations with caution, suggesting that caution can coexist with confidence in GenAI capabilities. We introduce the TiAI scale, an adapted instrument designed to measure calibrated reliance. These findings support a nuanced understanding of trust in GenAI, distinguishing between uncritical acceptance and a prudent, uncertainty-aware approach to generative outputs.