Artificial intelligence, especially large language models, has rapidly entered everyday life, yet public attention to their environmental footprint has lagged behind discussions of capability, ethics, and governance. Addressing this visibility gap, this study develops a three-level computational framework grounded in ecolinguistics environmental communication to examine how TikTok discourse constructs GenAI's environmental impact. We analyzed related TikTok videos with a multi-method pipeline integrating techniques in computational linguistics and corpus linguistics, including semantic network, linguistic feature analysis, topic-modeling, concordance, and sentiment analysis. Results indicate a strongly asymmetric action structure in which technological and infrastructural entities are positioned as dominant agents, while environmental entities appear primarily as affected resources. Verbs linked to environmental objects are dominated by “ use ” and other harm-oriented actions, with few mitigation-oriented relations. The topic hierarchy is centered on AI's environmental burdens, with only limited attention to benefit frames. Responsibility discourse frequently mentions necessity and obligation, and the emotional profile is heavily skewed toward fear. The framework demonstrates how ecolinguistic concerns can be operationalized at scale while remaining interpretable through corpus-based contextual inspection. Codes and data are available on GitHub 1 .
Qinghao Guan (Wed,) studied this question.