Visuals such as images, charts, and diagrams are widely used in sustainability reports to convey key messages, with colour—particularly green—often employed to evoke emotional connections to environmental themes. While colour can significantly influence readers, its analysis in academic research remains challenging due to subjectivity and limited sample sizes. This study demonstrates how generative AI, specifically large language models (LLMs) like ChatGPT, can be used to evaluate the use of colour in sustainability reports at scale. By mimicking human cognitive functions while maintaining objectivity, AI enables more accurate and efficient analysis across larger datasets than traditional methods allow. We present a worked example that integrates text and visual analysis, showing how AI models can be prompted, fine-tuned, and enhanced with domain-specific knowledge to support and augment human interpretation. Our approach mitigates both human and AI biases and improves reliability compared to zero-shot applications. The paper contributes a research protocol for AI-assisted colour evaluation, offering a scalable and transferable method for future research. We discuss implications for researchers, practitioners, and policymakers, and highlight how AI can enhance both the methodological rigour and practical relevance of accounting research.