The increasing use of artificial intelligence (AI) in language learning has introduced new opportunities for speaking practice through automated speech evaluation systems. However, little is known about how learners experience AI-generated feedback and perceive their speaking performance development over time. This study aimed to explore EFL learners’ experiences in speaking practice using Moodie.ai, focusing on their responses to AI feedback and their perceived speaking performance development. This study employed a qualitative narrative inquiry design. Three undergraduate EFL learners who had used Moodie.ai for speaking practice participated in the study. Data were collected through semi-structured interviews, narrative frames, and artifacts, and were analysed using reflexive thematic analysis. The findings revealed that learners demonstrated interconnected cognitive, affective, and behavioral responses toward Moodie.ai feedback. Cognitively, they developed greater self-awareness, identified areas requiring improvement, expanded their understanding of effective speaking, and critically evaluated the credibility of AI feedback. Affectively, they experienced encouragement, curiosity, confusion, frustration, and varying levels of confidence. Behaviorally, they applied feedback, engaged in repeated speaking practice, and verified AI feedback through self-evaluation and external sources such as lecturers, peers, and dictionaries. Learners also perceived low speaking performance development. Greater improvement was reported in fluency, vocabulary, body language, facial expressions, and gestures, whereas pronunciation, grammar, and comprehension showed more limited development. Overall, the findings suggest that learners' experiences with Moodie.ai primarily involved reflecting on AI-generated feedback, interpreting its usefulness, and responding through subsequent speaking practice, while learners remained the central agents in shaping their own learning experiences.