<![CDATA[Background: In nations like Indonesia that are vulnerable to natural disasters, early warning systems are essential. Warning messages serve as institutional speech acts in addition to providing meteorological information, with the communicative function of influencing public behavior and risk perception. However, whether these intended functions translate into actual audience effects is an empirical question outside the purview of this textual study. The purpose of this study is to investigate the linguistic realization of speech act configurations and design techniques in BMKG online weather early warning messages. Methodology: The analysis uses a corpus of 120 online weather early warning texts produced by BMKG, Indonesia's national meteorological office, between November 2025 and February 2026, using a qualitative corpus-assisted design with descriptive quantitative mapping. Findings: The results show the forms of BMKG warnings are very uniform with the integration of the directive, predictive and representational speech acts. The use of language for institutional attribution lends the project authority and credibility, but temporal deixis, modality, intensity, and spatial definition also build urgency. 93.3% of the corpus contained a seven-move structure that was found repeatedly, with content pieces changing systematically depending on forecast confidence and danger categorization. Conclusion: The study shows that the BMKG’s online alerts are a stable institutional genre in which the pragmatic strategy and linguistic form are always in accordance with the public risk communication. The data offer a textual baseline for further investigation of audience reception and behavioural response. Originality: This study contributes to pragmatics and corpus-assisted discourse analysis by providing empirical evidence of the realisation of institutional speech acts in digitally mediated early warning communication in a multilingual setting.]]>
