Comparing instructor and role-play AI chatbots for EFL writing: effects on anxiety, engagement, and feedback perception

Grounded in foreign language anxiety (FLA) research and second-language acquisition theories, this mixed methods study investigates how two pedagogically distinct artificial intelligence (AI) GPT-4o-powered chatbots influence adult English as a Foreign Language (EFL) learners’ writing anxiety, engagement, feedback perception, and future-use preferences. In a counterbalanced within-subjects design, 21 learners each completed two 15-minute writing sessions: one with an Instructor Bot providing explicit corrective feedback, and another with a Role-Player Bot delivering implicit, scenario-based conversational guidance. Quantitative analyses showed that the Role-Player Bot significantly reduced learners’ mistake-related anxiety. At the same time, the Instructor Bot’s explicit feedback was rated as significantly more useful and confidence-boosting. Engagement and enjoyment were high and comparable across conditions; however, participants’ preferences for future chatbot use varied based on their learning goals, which were either accuracy-oriented or fluency-oriented. Qualitative thematic analysis revealed a clear preference for hybrid chatbot designs combining conversational immersion with structured, explicit corrections. These results underscore key trade-offs between accuracy-focused and narrative-driven AI tutors, highlighting the importance of adaptive chatbot architectures that dynamically balance explicit feedback and engaging, low-pressure interaction. The study provides novel insights into chatbot-assisted writing support and offers actionable guidelines for designing personalized, anxiety-sensitive AI tools in EFL contexts.