Transforming uncertainty into exploration: effects of text-to-image AI on EFL learners' writing anxiety, motivation, and performance

Abstract

English as a Foreign Language (EFL) writing is inherently fraught with linguistic uncertainty, which frequently triggers anxiety and stifles learner motivation. While generative AI (GenAI) is increasingly adopted, current research predominantly treats these tools as text-generation shortcuts for final writing products, leaving the cognitive and affective mechanisms of multimodal GenAI as a process-oriented scaffold largely underexplored. To bridge this gap, this study investigated the effects of a text-to-image (T2I) AI-supported visual narrative intervention on 78 EFL undergraduates' writing anxiety, motivation, and performance through an 8-week quasi-experiment. Specifically, the intervention required the experimental group to engage in an iterative cycle of text input, image generation, and prompt refinement to construct their drafts, whereas the control group employed traditional text-only drafting. Quantitative results revealed that the experimental group exhibited significantly lower writing anxiety, alongside enhanced situational writing motivation and a modest but significant improvement in writing performance, compared to the control group. Qualitative thematic analysis of retrospective interviews indicated that, during the intervention, the T2I AI transformed linguistic uncertainty from a psychological threat into a gamified exploration, although the durability of this shift beyond the intervention period remains to be established. The immediate, non-judgmental visual feedback functioned as an autonomy-supportive scaffold, driving learners to notice the gap and autonomously engage in lexical risk-taking and syntactic refinement. However, as the intervention bundled visual feedback with AI immediacy and interactivity, the specific contribution of the visual modality warrants further investigation. This study contributes to reframing the understanding of uncertainty in GenAI-supported EFL writing and highlights the pedagogical affordances of leveraging multimodal AI as a collaborative, affective scaffold rather than a mere linguistic shortcut.