This study investigates the hermeneutic potential of artificial intelligence in lemmatic analysis–a method that reduces word forms to canonical lemmas and, in its extended version, identifies stable semantic patterns that form the conceptual framework of a text. The article tests the hypothesis that AI can perform not only an auxiliary but also a heuristic function, revealing implicit semantic structures inaccessible to traditional methods. Based on two previously manually analysed, rather disparate corpora – Friedrich Nietzsche’s metatexts (personal correspondence, drafts, notes) and Harold Robbins’s popular novels – the study presents a three-level model of AI-assisted lemmatic analysis. Level 1 automates lemmatization and frequency counting; Level 2 identifies stable lemmatic patterns and binary oppositions; Level 3 integrates these results into a hermeneutic interpretation. To validate the model, an experiment compares AI-generated lemmatic analysis (using GPT-4) with manual results. The experiment shows that AI successfully replicates core-periphery dichotomies and detects instrumental functions of thematic lemmas (e.g., money as a means of escape in Robbins), but may miss certain peripheral lemmas (e.g., “wound”) identified by human researchers. The article introduces the concept of “hermeneutic AI” – an algorithm capable of dialogical, traceable, and verifiable interpretation – and proposes five validation criteria: traceability, pattern stability, inter-coder verification, completeness, and heuristic productivity. The findings demonstrate that AI-assisted lemmatic analysis offers scalable, reproducible tools for digital humanities while remaining complementary to human expertise. The study contributes to the methodology of text interpretation and opens new perspectives for the integration of machine learning into literary and philosophical hermeneutics.