We use high-frequency Google search data, combined with data on the announcement dates of non-pharmaceutical interventions (NPIs) during the COVID-19 pandemic in U.S. states, to disentangle the short-run direct impacts of multiple different state-level NPIs in an event study framework. Exploiting differential timing in the announcements of restaurant and bar limitations, non-essential business closures, stay-at-home orders, large-gatherings bans, school closures, and emergency declarations, we leverage the high-frequency search data to separately identify the effects of multiple NPIs that were introduced around the same time. We then describe a set of assumptions under which proxy outcomes can be used to estimate a causal parameter of interest when data on the outcome of interest are limited. Using this method, we quantify the share of overall growth in unemployment during the COVID-19 pandemic that was directly due to each of these state-level NPIs. We find that between March 14 and 28, restaurant and bar limitations and non-essential business closures can explain 6.0% and 6.4% of UI claims respectively, while the other NPIs did not directly increase own-state UI claims. This suggests that most of the short-run increase in UI claims during the pandemic was likely due to other factors, including declines in consumer demand, local policies, and policies implemented by private firms and institutions.