<![CDATA[Background: Mental health chatbots are increasingly positioned as accessible support tools in Indonesia, yet their capacity to provide care depends not only on technical responsiveness but also on relational language, emotional safety, and cultural-pragmatic fit. Objective: This study aims to evaluate how Indonesian mental health chatbot discourse constructs empathy, manages emotional risk, and aligns with local communicative norms of help-seeking. Method: Using a qualitative discourse-pragmatic design, this study analysed 23 public text units from chatbot-related platforms, institutional pages, government health resources, and scholarly documents through a relational empathy matrix, emotional safety audit, and cultural-pragmatic alignment checklist. Results: Findings show that empathy is most visible through relational presence and autonomy support, while affective recognition, validation, and explicit non-judgmental stance appear less consistently. Emotional safety is strongest when chatbot discourse foregrounds boundary transparency, diagnosis restraint, and human-in-the-loop referral, but weaker when crisis wording, refusal style, and consent framing remain underdeveloped. Implication: Cultural alignment is supported by accessible Indonesian and locally credible referral pathways, although stigma, family pressure, multilingual diversity, and regional variation receive limited explicit attention. Novelty: This study offers a novel framework for evaluating mental health chatbot empathy as a relationally expressed, emotionally bounded, and culturally situated design practice.]]>
