This paper introduces a new framework for studying the moral history of the novel through the lens of large language models (LLMs). Drawing on over 9,000 Wikipedia plot summaries of 20th- and 21st-century novels, it demonstrates how LLMs can surface the implicit life lessons – or story morals – encoded in narrative summaries at scale. Building on recent work in moral inference and narrative abstraction, the study proposes a reflexive, perspectival approach that emphasizes interpretation over taxonomy. To account for the semantic variability of LLM-generated morals, the study employs a randomized prompt assignment strategy and analyzes the resulting moral keywords using co-occurrence networks and hierarchical clustering, enabling the identification of latent moral communities and comparison across modeling approaches and time. Taken together, the findings argue for the value of LLMs not only in extracting narrative values, but in enabling a new, culturally situated view of literary history through computational means.