This chapter explores the application of Large Language Models (LLMs) for identifying emotional appeals (pathos) in natural language argumentation. Building on the interactional model of pathos presented earlier in the book (MIPA), it demonstrates how LLMs can support scalable annotation of rhetorical emotional appeals across languages and genres. The first study applies GPT-4 to competitive debate transcripts, analyzing categorical emotions (anger, fear, joy, sadness, disgust) and comparing results with manual annotation. The second study investigates multilingual political discourse in six European countries during pre-election debates, focusing on the prevalence of fear appeals in regions near armed conflict. In both cases, LLMs demonstrate fair intersubjective agreement with human annotation, though limitations remain in context sensitivity and mixed-emotion detection. The chapter argues that, despite their computational nature, LLMs can emulate annotation tasks when equipped with carefully designed prompts and evaluation metrics. Methodological reflections emphasize the need for hybrid approaches that combine computational and qualitative insights. The study confirms that pathos can be systematically mined using generative AI, offering a new avenue for interdisciplinary research in the areas of argumentation, computational linguistics, and emotion studies.