Enhancing human-AI communication in language education: Developing customized chatbots for teaching genre analysis and annotation

Generative artificial intelligence (GenAI) holds significant promise for transforming language education, yet its application for advancing academic writing skills—such as genre analysis in English for Research Publication Purposes (ERPP)—remains underexplored. This study investigates how system-level customized GenAI chatbots, developed using an in-house chatbot customization platform, can be systematically designed to scaffold the teaching and learning of genre analysis and annotation for research students. Guided by the DEER praxis framework, we iteratively refined chatbot system prompts and workflows across four instructional stages: background survey, teaching/review, practice, and reflection. Two chatbot versions were deployed with eleven doctoral students from diverse disciplines at a Hong Kong university. Analyses of chat histories and embedded questionnaires revealed that the customized chatbot provided scalable, interactive support for move-step annotation tasks, elicited reflective engagement, and delivered adaptive feedback at each stage, with iterative refinements producing smoother user interaction and more coherent learning processes. Key strategies for success included stage-specific prompt engineering, dynamic follow-up questioning, and continuous human oversight and reflection. Our findings illustrate that customized AI tools can complement genre pedagogy when integrated with explicit instructional design and iterative refinement. By documentinging a traceable process of system-level chatbot customization, this study contributes practical insights to technology-mediated language education and extends understanding of human–GenAI communication in specialized academic contexts.