With global warming intensifying, average citizens are beginning to experience its impact more directly. More than 100 recent studies have examined the link between direct extreme weather experience and shifts in climate policy preferences but have found inconclusive results. One reason for this inconclusiveness is that so far, it is not well understood what average citizens perceive as extreme weather. This project develops a novel framework to analyze how average people perceive extreme weather based on open-ended survey responses. We leverage various automated text-as-data methods to analyze open-ended responses across four languages in a geocoded Swiss panel study based on a probability sample. Based on structural topic models, transformer-based classifiers and a novel zero-shot classification method, we capture in depth when and how people perceive extreme weather in the context of climate change (for example, based on precipitation or temperatures) and to distinguish whether they pay more attention to sudden-onset or slow-onset types of extreme weather. Based on respondents’ geolocations, we link their subjective perceptions to objective weather data to see to what extent the self-described extreme weather phenomena are aligned with objective local measures of extreme weather and whether this hinges on characteristics of their surroundings (e.g., share of local green spaces). We contribute to the question of whether extreme weather experience has the potential to increase preferences for stricter climate policies by clarifying what laypeople perceive as extreme weather. In addition, we hope to advance best practices on data integration of subjective and objective measures of environmental data.