A new study introduces a framework that analyzes how prompting influences the internal representations of large language and vision-language models, revealing that prompts significantly reshape model behavior. The findings suggest that understanding these geometric transformations can enhance the effectiveness of AI models in various tasks.