AI explainability and employee acceptance of generative AI at work: The mediating role of trust in AI
Year: 2026
Type: article
Abstract: This study develops and tests an employee-centered model examining how perceived AI explainability is associated with acceptance of generative artificial intelligence (GenAI) at work. It distinguishes three practical dimensions of explainability, including transparency, interpretability, and actionability, and examines whether trust in AI statistically mediates the associations between these explanation qualities and employees' willingness to use, rely on, and recommend GenAI in organizational tasks. This study employed a cross-sectional explanatory mixed-method design with 486 employees in China and 18 completed semi-structured interviews. Twenty-two seven-point Likert items were analyzed in SmartPLS 4 using reliability and validity assessment, structural path analysis, 5,000 bootstrap samples, effect sizes, and predictive diagnostics. The estimates indicate that transparency (beta = .249), interpretability (beta = .300), and actionability (beta = .292) are positively associated with trust in AI. Trust is strongly associated with employee acceptance (beta = .518) and statistically mediates the associations between all three explainability dimensions and acceptance, with the direct paths remaining significant. The model explains 48.8% of the variance in trust and 58.1% of the variance in acceptance. Interview findings further show that employees value explanations most when they reveal uncertainty, support verification, and preserve meaningful human control. These results represent statistical associations, not evidence of temporal or causal mediation. The study advances workplace GenAI research by decomposing explainability into employee-relevant affordances and positioning calibrated trust as a mediating pathway between explanation quality and acceptance. The model and reviewed literature suggest that explanation design should not be evaluated by explanation volume alone; organizations should provide layered explanations that disclose limitations, make outputs comprehensible, and help employees check, correct, and appropriately contest AI outputs.
Source: PaperAsia
Author: Zongyu Li
Institution: The University of Queensland
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7225752093
FWCI:
Cites:
Cited by:
Related to: