Abstract: Climate change is one of the defining challenges of our time, but just as important is how we talk about it: who shapes the conversation, which languages are included, and whose experiences are represented. In South Africa, climate change discourse in public media has been analysed almost exclusively in English, leaving a large part of the national conversation outside the evidence base used to shape policy. This study examines how climate change is framed across English- and isiZulu-language media sources, using a corpus of 186 documents from eight South African outlets. Applying a five-theme discourse taxonomy and zero-shot classification across two AI language models, a generic English model and an Africa-centric multilingual model, the study identifies systematic asymmetries in how the two languages frame climate change. English discourse prioritises economic development and formal policy debate; isiZulu centres community experience and lived environmental risk. Model benchmarking reveals that the Africa-centric model nearly doubles the generic model's classification performance on isiZulu documents, confirming that standard English AI tools systematically underrepresent African-language climate narratives.