The study investigates how conversational AI chatbots, such as ChatGPT and Replika, support Pakistani ESL learners in developing oral fluency and intercultural communication skills. Complex dynamic systems theory, Krashen's affective filter hypothesis, and Bandura's social cognitive theory suggest that AI-mediated practice reduces anxiety and improves speaking outcomes by enhancing technical self-efficacy. To test hypotheses on chatbot usage, interaction type, and significant learner variables, multiple and hierarchical regression were applied to the dataset. Structured, role-playbased AI interaction predicts oral fluency and communicative skill better than formal training. Chatbot use and communication anxiety reduction are highly moderated by technological self-efficacy. The benefits demand internet access and self-efficacy, highlighting unfairness risks. The study found that AI chatbots are effective instructional aids. AI conversation practice in curricula requires planned, scaffolded integration, proactive development of learners' digital and linguistic self-efficacy, and policy action to reduce the digital gap. This evidence-based method is needed to realize AI's egalitarian and revolutionary potential in Pakistan's ESL setting.