Instruction given to an LLM that sets its role, behavior, or output format before any user input arrives. Acts as the model's standing operating instructions for the rest of a conversation. Wrapper tools usually contain a hidden system prompt you cannot see.
Definition: Instruction given to an LLM that sets its role, behavior, or output format before any user input arrives. Acts as the model's standing operating instructions for the rest of a conversation. Wrapper tools usually contain a hidden system prompt you cannot see.
A system prompt is the model's standing instructions, set before any user message. It can pin a role ("you are a research assistant"), enforce constraints ("never invent quotes"), specify output formats ("respond in JSON"), or load context the model should always have access to. The user does not normally see the system prompt.
For research transparency, the system prompt is consequential. It shapes every output the model produces, but wrapper tools hide their system prompts as proprietary IP. Two tools using the same underlying LLM can produce wildly different results because of differently engineered system prompts. Without visibility into the prompt, comparison across tools is unreliable.
Direct API access exposes the system prompt to the developer (you write it). Wrapper tools usually do not. This is one of the core reasons rigorous research workflows favour API-first over wrapper-tool dependence.
This term is referenced in the following articles: