System Prompt vs User Prompt

System prompts and user prompts serve different purposes in an AI application.

Start from the task

A system prompt contains high-level instructions that define how the model should behave. It might describe the application role, output rules, safety constraints, business policies, or formatting requirements.

For example:

Make the prompt testable

“You are a customer-support assistant. Answer only using the supplied knowledge base. If the answer is unavailable, say you do not know.”

A user prompt represents the request or information supplied by the person using the application:

Measure before deployment

“How can I cancel my subscription?”

Separating these responsibilities makes AI applications easier to maintain.

Practical checks before you proceed

Stable behavior belongs in higher-priority application instructions, while changing requests belong in user messages. Developers should not rely on users to repeat important application rules every time.

Both kinds of content generally contribute to the request's token usage. A 5,000-token system prompt can therefore become expensive when repeatedly sent across millions of API calls.

Practical checks before you proceed

This is why system prompts should be clear but not unnecessarily long.

Avoid placing an entire company knowledge base directly in the system prompt. Keep core instructions there and retrieve relevant knowledge separately.

Practical checks before you proceed

Also remember that system prompts are not secret storage. Never place passwords, private credentials, or API secrets inside prompts simply because users cannot directly see them in the interface.

A good prompt hierarchy makes responsibilities clear: the application defines behavior, the user defines their request, and retrieved context supplies the information needed to answer it.