Abstract
Artificial intelligence (AI) is transforming language assessment, yet the psychological mechanisms through which it influences complex communicative skills remain underexplored. This study examines how AI-augmented online assessment (OA) relates to self-perceived English-speaking competency (ESC) among vocational English-as-a-Foreign-Language (EFL) learners in Guangdong Province, China, focusing on the mediating roles of motivation in digital teaching (MODT) and engagement in digital teaching (ENDT). Data from 463 students across three public vocational colleges were analyzed using Partial Least Squares Structural Equation Modeling (PLS-SEM) and fuzzy-set Qualitative Comparative Analysis (fsQCA). Results indicate that OA has a small but significant direct association with ESC, while its indirect effects via motivation and engagement are substantially stronger. A marker-variable analysis suggests that common-method variance modestly inflates some direct paths but does not alter the overall pattern. fsQCA identifies several sufficient configurations for high ESC, including the concurrent presence of OA, motivation, and engagement; high motivation with OA even under lower engagement; and strong motivation and engagement even in the absence of OA. Across all configurations, motivation consistently emerges as the core condition, underscoring its central role in sustaining performance and perceived language development. Out-of-sample prediction (PLSpredict) confirms that the model most accurately predicts fluency and coherence-the sub-skills best captured by AI feedback-while prediction for vocabulary and rhetorical expression is weaker. Overall, the findings clarify how Learning-Oriented Assessment operates within AI-enabled vocational contexts, highlighting that feedback effectiveness depends less on automation than on perceived credibility, competence enhancement, and vocational relevance.