From Social Risk to Digital Sanctuary: EFL Learners’ Perceptions of Emotionally Neutral AI in Oral Language Practice

Abstract

As generative artificial intelligence (GenAI) becomes increasingly integrated into second language (L2) education, its affective implications remain underexamined. This mixed‑methods study investigates how 47 Iranian university‑level English as a Foreign Language (EFL) learners perceive AI‑based speaking tools in relation to anxiety, communicative confidence, and emotional openness. Drawing on Krashen’s Affective Filter Hypothesis (AFH), the Willingness to Communicate (WTC) model, and the Computers Are Social Actors (CASA) framework, the study examines learner perceptions based on their ongoing, naturally occurring use of AI‑mediated speaking tools. Descriptive quantitative findings indicate that participants perceive AI interaction as involving reduced evaluative pressure and increased willingness to experiment linguistically. Qualitative responses further indicate that learners view AI as a non‑judgmental interlocutor that supports emotional openness and low‑stakes self‑disclosure. Rather than positioning AI as a replacement for human interaction, the findings suggest that learners view AI tools as preparatory rehearsal spaces that may support confidence development. This study extends existing research on affect, anxiety, and mediated communication in SLA by examining these constructs in AI‑mediated speaking contexts.