Libyan EFL Students’ Perceptions of AI-Driven Tools for Developing English Speaking Skills: A Study at the Faculty of Arts and Sciences, Suluq, Benghazi University

Abstract

This exploratory mixed-methods study examines perceptions of AI speaking tools among 42 Libyan EFL undergraduates. Quantitative questionnaire data (Cronbach’s .89) were triangulated with qualitative thematic analysis. Results indicate positive perceptions for fluency (3.69), grammatical accuracy (3.58), and communicative confidence domain (3.62), alongside strong overall endorsement (4.07). Although learners felt confident using AI-practiced phrases in real settings, this confidence was less robust in spontaneous interactions (3.38). Over-correction emerged as a notable qualitative concern. Correlational analysis revealed moderate positive relationships between duration of AI usage and perceived fluency (.448, p = .003}), structural accuracy (.382, p = .012}), and communicative confidence ( .415, p = .006), suggesting that sustained engagement with AI tools is associated with more positive perceptions across all three speaking dimensions. Grounded in Krashen’s Affective Filter, Swain’s Output, and Long’s Interaction hypotheses, findings suggest AI effectively scaffolds form-focused practice and reduces anxiety but cannot replicate the dynamic negotiation of meaning essential for full communicative competence.