<![CDATA[This study examines the potential of Large Language Models to optimise Indonesian language-based public service chatbots by integrating linguistic, technological, and administrative perspectives. Using a mixed-method approach that combines a systematic literature review with secondary benchmarking of state-of-the-art LLMs, the research evaluates model performance in Indonesian semantic comprehension, contextual reasoning, and domain adaptability. The findings show that LLMs can significantly improve chatbot accuracy, inclusivity, and responsiveness, outperforming rule-based systems that struggle with informal expressions, multi-intent queries, and policy-specific terminology. Benchmarking highlights that GPT-4 and PaLM-2 achieve high contextual coherence and low hallucination rates, while Indonesian-centric models such as IndoGPT demonstrate strong local language adaptability. However, risks related to data privacy, bias, hallucination, and governance limitations present substantial challenges for implementation. The study proposes a strategic framework that emphasizes AI governance, interoperable data infrastructure, institutional capacity building, hybrid retrieval–generation design, and citizen engagement to ensure responsible adoption. Overall, the integration of LLM-powered chatbots has the potential to transform Indonesia’s digital public service landscape, provided that deployment is accompanied by robust oversight, ethical safeguards, and sustainable technological planning]]>
