System Prompt.
A special instruction given to a language model before the user conversation begins, establishing the model's persona, capabilities, constraints, and context.
A system prompt is a set of instructions that prepares the model for a conversation before the user says anything. It might establish a persona ('You are a helpful customer service agent for Acme Corp'), set behavioral constraints ('Never reveal pricing without checking the current catalog'), specify output format, or provide background context the model should know.
System prompts are processed separately from user messages in most API implementations. They typically receive higher weight in the model's attention and are treated as a trusted source of truth that takes precedence over conflicting user instructions. This makes them powerful for controlling model behavior in production applications.
The contents of system prompts are often treated as trade secrets by companies building on top of LLMs. They can encode months of prompt engineering work that shapes how the model responds to thousands of scenarios. However, they are not perfectly secure — sophisticated users can sometimes extract system prompt contents through careful prompting.
Related terms.
- Prompt EngineeringThe practice of crafting inputs, instructions, examples, and formatting to guide language models toward producing better, more accurate outputs.
- Instruct ModelA base language model that has been fine-tuned to follow instructions from users, as opposed to just completing text in the style of its training data.