While AI chatbots are increasingly integrated into language education, most studies have focused on student perceptions rather than objectively measuring linguistic improvement (Chen et al., 2023; García-Sánchez & Pérez-Paredes, 2024; Kohnke & Zou, 2023). This study presents a longitudinal analysis of chatbot interactions among Japanese university students (n=18) to examine engagement patterns, linguistic complexity, grammatical accuracy, and self-correction behaviors over an academic term. Using statistical analysis techniques, including Correlation Analysis, Paired t-tests, and ANOVA, key findings reveal that students demonstrated statistically significant improvements in mean sentence length, response expansion, and self-correction awareness (p < 0.05). However, lexical diversity remained unchanged, suggesting that students relied on familiar vocabulary despite producing longer, more structured sentences. Additionally, interaction patterns, such as total turns per session and follow-up question rates, showed no significant trends over time. These results suggest that AI chatbots can effectively encourage longer, more structured responses and increased self-correction habits, but may require additional pedagogical interventions to enhance vocabulary development. This presentation will also discuss practical implementation strategies to optimize chatbot-based learning for linguistic development.