There has been considerable speculation regarding how people cope during the COVID-19 pandemic; however, surveys requiring selection from prespecified answers are limited by researcher views and may overlook the most effective measures. Here, we apply an unbiased approach that learns from people's collective lived experiences through the application of natural-language processing of their free-text reports. At the peak of the first lockdown in the United Kingdom, 51 113 individuals provided free-text responses regarding self-perceived positive and negative impact of the pandemic, as well as the practical measures they had found helpful during this period. Latent Dirichlet Allocation identified, in an unconstrained data-driven manner, the most common impact and advice topics. We report that six negative topics and seven positive topics are optimal for capturing the different ways people reported being affected by the pandemic. Forty-five topics were required to optimally summarize the practical coping strategies that they recommended. General linear modelling showed that the prevalence of these topics covaried substantially with age. We propose that a wealth of coping measures may be distilled from the lived experiences of the general population. These may inform feasible individually tailored digital interventions that have relevance during and beyond the pandemic.