In the era of advancing artificial intelligence in education, this study investigates whether regular practice with AI-powered chatbots improves English speaking proficiency more effectively than traditional peer-to-peer conversation practice. Grounded in the interaction hypothesis (Long, 1996) and output hypothesis (Swain, 1985), which emphasize the role of interactive practice and feedback in language acquisition, the research employed an experimental design. Two groups of English learners participated over 12 weeks: an experimental group practiced speaking with Elsa AI, an AI chatbot equipped with natural language processing for real-time feedback on pronunciation, vocabulary, grammar, and fluency, while a control group engaged in peer-to-peer conversation practice. Speaking proficiency was assessed pre- and post-intervention using standardized tests (IELTS speaking), and statistical methods (e.g., t-tests or ANOVA) were used to compare proficiency gains. Statistical analyses revealed that both groups improved significantly over time, but the chatbot group achieved larger mean gains than the peer group. These findings suggest that AI chatbots offer scalable, on-demand practice with immediate corrective feedback, addressing limitations of human partners such as availability and consistency. The findings could inform the integration of chatbots in diverse English Language Teaching (ELT) settings, enhancing speaking practice opportunities globally.
