In a scenario where you need to improve the responses of an Azure OpenAI model for a legal document summarization task, what prompt engineering techniques would you apply to ensure the model generates accurate and concise summaries? Describe the steps you would take to refine the prompts.