7 Prompt Types
This chapter covers
- Understanding how these systems process structured conversations with distinct message types
- Recognizing the three message types: system, user, and assistant
- Designing system prompts that define what the AI does across all interactions
- Crafting user prompts that provide queries or context data for processing
- Applying a decision framework to decompose prompts into system and user components
- Determining when architectural separation provides value versus when to use simple prompts
You have built a comprehensive prompt that works perfectly. It contains 300 words of carefully crafted instructions defining role, output format, communication style, and task procedures, followed by the data you want processed. When you need to process different data, you duplicate the entire prompt and swap the content at the end. Instructions and input are bundled together as a single block of text, inseparable.
This is a monolithic prompt. It combines two concerns that should be separate: behavioral instructions that define what the AI does, and input data that varies with each request. Contemporary LLMs support decomposition through distinct message types. You define behavioral guidelines once in a system prompt, then provide variable input separately in user prompts. This creates clean architectural boundaries, instructions configure the system, data comes from users.