Optimizing prompts so repeated text is shared across API calls, reducing redundant token processing and costs.