Learner Perceptions and Interactional Features in ChatGPT-Mediated EFL Practice: A Comparative Study Across Proficiency Levels

This study investigates how AI-mediated dialogue can support English as a Foreign Language (EFL) learning by comparing the experiences of two learner groups (CEFR A2-B1 and B2-C1) interacting with ChatGPT 3.5. Combining corpus-based and perception data, the study analyses written interactions and post hoc questionnaire responses to explore four dimensions of learner experience: politeness, emotional comfort, perceived advantages and disadvantages, and conversational naturalness. Learners engaged in two tasks (small talk and role-play), and their interactions were analysed for topic preference, lexical variation, sentiment tone, and use of politeness markers. Perceptual data were collected via Likert-scale and open-ended items. Results suggest that while learners across both groups viewed the chatbot experience as generally positive, lower-proficiency learners reported greater emotional comfort, whereas higher-level learners were more critical of the chatbot’s conversational depth and naturalness. The findings contribute to ongoing research on AI in language education, offering evidence-based insights for designing adaptive, proficiency-sensitive chatbot interactions.