Trusting AI feedback in EFL writing: Anxiety, feedback literacy, and writing self-efficacy

This three-wave, nonexperimental study examined whether artificial intelligence (AI) feedback trust (AIFT) was associated with subsequent English as a foreign language academic writing self-efficacy (EAWSE) directly and indirectly through writing anxiety (WA), and whether second language writing feedback literacy (LWFL) and perceived AI transparency (PAIT) conditioned the first-stage association. AIFT was measured at Time 1; writing anxiety, second language writing feedback literacy, and perceived AI transparency were measured at Time 2; and EAWSE was measured at Time 3. The final matched sample comprised 534 Chinese university students with experience using artificial intelligence-generated writing feedback. Partial least squares structural equation modeling showed that AIFT was positively associated with EAWSE (β = 0.180, p = 0.002), inversely associated with WA (β = -0.178, p = 0.008), and indirectly associated with EAWSE through WA (β = 0.081, p = 0.009). The AIFT × LWFL interaction on WA was positive (β = 0.125, p < 0.001), indicating that higher LWFL attenuated, rather than strengthened, the negative AIFT-WA association; H5 was therefore not supported, although the interaction was statistically significant in the opposite direction. The AIFT × PAIT interaction was not significant (β = 0.040, p = 0.514). These findings position trust as a task-specific appraisal whose association with writing self-efficacy operates partly through affective experience and varies with learners' capacity to evaluate feedback. They also emphasize calibrated trust and critical feedback use rather than unconditional reliance on artificial intelligence.