In the late 1930s, a team of writers traveled the United States to record the life stories of thousands of older Americans for posterity. We combine close human readings with text analysis by large-language models (LLMs) to turn these unstructured texts into a rich data set that captures the narrators’ experiences, economic fortunes, and perspectives on what brought happiness, satisfaction, and meaning into their lives. We first demonstrate that, under the right circumstances, LLMs can detect and analyse factual information in this corpus in a comparable way to human readers. We then document a set of facts summarizing the determinants of life satisfaction for these narrators. We find that a substantial fraction of respondents derived satisfaction from their work and career, in contrast with a rich body of work in psychology that emphasizes the paramount importance of close personal relationships. Those that derived satisfaction from the bonds of family were much more likely to emphasize religion and education, and those migrating when young were far more likely to focus on the joys of adventures and satisfaction of being independent. Those with the lowest perceived life satisfaction dwelled much more on perceived negative shocks, such as medical conditions or having to drop out out school due to poverty.