AI-driven chatbots have emerged as promising tools for personalized learning and inclusive education. However, their practical use in foreign language learning remains underexplored, particularly regarding engagement patterns and barriers faced by students. This study examines student interactions with ‘Vulpy’, a chatbot integrated into an online learning platform, focusing on the extent to which its use aligns with its intended educational purpose. Using a mixed-methods approach, we conducted a quantitative content analysis of 7,810 student messages alongside an in-depth qualitative analysis. Findings reveal that a significant proportion of students did not use the chatbot for foreign language learning but instead engaged in social, experimental, or rule-breaking interactions. Identified barriers include linguistic, interactional, and motivational challenges, limiting its inclusive potential for learning. We discuss implications for adaptive chatbot design and structured pedagogical integration to foster equitable learning experiences. The study highlights the necessity of scaffolding mechanisms to support diverse learners effectively.