The recent global discourse on tariff wars has sparked widespread social media reactions. This study investigates how distinct narratives surrounding tariff policies propagate online by integrating large language models (LLMs) with epidemiological modeling. We collect and cluster tweets reacting to U.S. trade measures, employing GPT-4o with structured prompts to extract coherent narrative themes. Five dominant narratives emerge: satire, macro-economic alarm, domestic political critique, regional economic anxieties, and strategic retaliation. To assess virality, we apply classic epidemiological models (SIR, SIS, SIRS, SEIR, SEIZ), estimating transmission rates and error. Our findings reveal that strategic retaliation narratives demonstrate highest transmissibility with low error across models. This narrative’s appeal to policymakers, experts, and media amplifiers, along with real-time developments, contributes to sustained propagation. Conversely, emotionally charged or satirical narratives show higher error rates and lower consistency. By bridging computational social science, AI, and epidemiology, this research offers a framework to understand how international economic discourse spreads on social media platforms.