Abstract: This project aims to investigate climate change discourse and policy perspectives across the SADC region using a combination of policy documents, academic research articles, and community and stakeholder insights from interview video transcripts. Different language models are used to analyse this data in order to identify structural biases such as colonial framing, external dependency framing, universalist assumptions, local knowledge omission, and differential impact recognition. The results offer a benchmarking analysis on the adequacy of both generic (ChatGPT and Deepseek), African-centric LLMs (AfroLM) and manual annotation in identifying colonial approaches to policy making, and will provide new insights to tackle climate change by acting on the public perception of the topic.