AI-assisted speaking in ELT: a bibliometric mapping of research gaps and a future research agenda (2023–2026)

The rapid integration of artificial intelligence (AI) into English Language Teaching (ELT) has generated a fragmented research landscape regarding AI-assisted speaking instruction. While individual studies report promising outcomes for chatbots and automatic speech recognition (ASR) systems, the field lacks a systematic mapping of trends, gaps, and future directions specific to ELT speaking contexts. This study addresses this gap through a bibliometric analysis of 251 Scopus-indexed articles (2023–2026), complemented by a supplementary review of 42 ERIC-indexed studies. Using VOSviewer for keyword co-occurrence analysis, we determined four research clusters: technical infrastructure (ASR, pronunciation, EFL), AI-ELT core, chatbot-mediated affective mediation (chatbot, anxiety, WTC), and pedagogical-affective factors (speaking, motivation, feedback). Overlay visualization revealed a temporal shift from technology-centric toward learner-centered, affect-sensitive research. Building on these findings, we propose a future research agenda comprising 15 research questions from the durability of AI-mediated affective gains to the effectiveness of ASR tools across diverse L1 backgrounds. To address the neglect of affective and ethical dimensions in AI-ELT teacher preparation, we introduce TPACK-A, an extension of the TPACK framework incorporating Affective Knowledge (AK) as a distinct teacher competency. This study provides researchers, practitioners, and policymakers with an actionable roadmap for advancing AI-assisted speaking instruction in ELT.