Protecting lives and livelihoods during volcanic eruptions is the key challenge in volcanology, conducted primarily by volcano monitoring and emergency management organisations, but it is complicated by scarce knowledge of how communities respond in times of crisis. Social sensing is a rapidly developing practice that can be adapted for volcanology. Here we use social sensing of Twitter (currently known as X) posts to track changes in social action and reaction throughout the 2018 eruption of Kīlauea on the island of Hawai`i. The volume of relevant posts very rapidly increases in early May, coincident with the beginning of the eruption; automated sentiment analysis shows a simultaneous shift towards more negative emotions being expressed in post text. Substantial negative trends in sentiment are evident in reaction to high-impact events, including the destruction of a popular residential area and injuries sustained by tourists viewing the eruption. Topics of local Twitter conversation reveal societal actions, including the sharing of hazard warnings, mitigation actions, and aid announcements. Temporal trends in societal actions reflect patterns in volcanic activity (e.g. the peak and waning of eruptive activity), civil protection actions (e.g. risk mitigation actions and the communication of official warnings), and socioeconomic pressures (e.g. the destruction of homes). Local tweets detailing eruption damage and disruption display a similar temporal trend to independent estimates of the number of buildings in contact with lava. We show how hazard and risk information is discussed and reacted to on Twitter, which helps inform our understanding of community response actions and aids situational awareness, and outline how our approach could be adapted for use in real time.