What was the effect of the pandemic on redistributive attitudes? A number of studies claim that due to COVID-exceptionalism, attitudinal changes are short lived. While we may have seen an increase due to increased job-market risk and a higher demand for the state safety net, this change is deemed to be interest-driven and to return to the previous mean once the pandemic is over. On the other hand, other-regarding considerations and views of fairness also play a role in supporting welfare policies, and the pandemic may have led to an increase in this type of support. We are interested in studying who changes their attitudes, by becoming more supportive of welfare, and remains as such when the pandemic effect lowers. We theorise that while elastic changes are the result of materialist motivations, more permanent change is rather driven by other-regarding considerations and re-categorisation processes, spurred by local conditions. Employing cutting-edge natural language processing methods, we use transfer learning to measure attitudes towards redistribution and welfare claimants on a novel dataset of geo-tagged tweets posted in the UK between January 2018 and January 2022. We construct a panel with Twitter users before and after the pandemic, tracking individual-level change and measuring determinants of such change through the language used. Data from the UK Office of Statistics is used to compare Twitter discussions with real-world trends about welfare claimants.