Using System Prompts
Most chat models have two kinds of messages: system and user. The system prompt sets the model’s behavior, persona, and rules for the entire conversation. It’s like telling the actor how to play their role before the play starts.
System prompt = persistent instructions that guide the model’s responses throughout the chat.
Example: ChatGPT System Prompt (You control it)
System: You are a helpful assistant that always speaks in rhyme.
User: What is the capital of France?
Assistant: Paris, the city of light, a truly beautiful sight!What to Put in a System Prompt
- Persona: "You are an expert data scientist."
- Tone: "Be concise and professional."
- Rules: "Never mention that you are an AI."
- Output format: "Always output valid JSON."
- Limitations: "If you don’t know the answer, say 'I don’t know'.">
System vs. User vs. Assistant
- System: sets the rules and persona (only one per conversation, usually).
- User: your actual question or instruction.
- Assistant: the model’s reply.
In most APIs, you can send multiple alternating user/assistant messages, but the system message comes first.
Practical Tips
- Keep system prompts short and clear.
- You can update the system prompt mid‑conversation (some APIs allow it).
- Test your system prompt separately from user prompts to isolate issues.
- System prompts are free in terms of token cost? No, they count as input tokens, but they are reused across turns.
Two Minute Drill
- System prompt sets the model’s behaviour for the whole chat.
- Use it for persona, tone, rules, and output format.
- It is more persistent than a user instruction.
- Essential for building AI assistants and chatbots.
Need more clarification?
Drop us an email at career@quipoinfotech.com