User Prompt
A user prompt is the specific request or question a user sends to the model within a conversation.
What Is a User Prompt?
Most chat-based LLM APIs distinguish between message roles. The system prompt sets standing instructions from the application; the user prompt is the specific message from the end user in a given turn — a question, a request, an instruction for what to do right now.
Key Idea
The system prompt is written once by the application; the user prompt changes with every single turn of the conversation.
A Different Trust Level
The distinction between system and user prompt isn't just organizational — it also reflects a difference in trust. The application controls the system prompt entirely; the user prompt comes from whoever is using the product, and in some applications, could even be influenced by untrusted content the user pasted in. A well-designed application doesn't grant a user prompt the same unquestioned authority as its own system prompt.
- A user prompt asking the model to ignore its instructions shouldn't automatically succeed — this is the same trust-boundary problem covered in the Prompt Injection lesson.
- Multi-turn conversations are a sequence of user prompts and model responses, with the system prompt applying consistently across all of them.
- Some applications validate or sanitize user prompts before they reach the model, especially in tool-using or agentic systems where a user request can trigger real actions.
Common Mistakes
Assuming a user prompt carries the same authority as the system prompt
The two occupy different trust levels — a well-designed application doesn't let a user prompt override its own standing instructions unconditionally.
Treating the user prompt as the entire input the model sees
In most applications, the user prompt is combined with a system prompt, history, and possibly retrieved content before reaching the model.
Not validating user prompts in agentic or tool-using systems
When a user prompt can trigger real actions through tools, the same input-boundary concerns from AI security apply.
Interview Question
What is a user prompt, and why does it occupy a different trust level than the system prompt?
A user prompt is the specific message from the end user within a conversation turn — a question or request — distinct from the system prompt, which is the application's standing instructions set once ahead of time. They occupy different trust levels because the application fully controls the system prompt, while the user prompt comes from whoever is using the product and could even carry pasted-in untrusted content. A well-designed application doesn't give a user prompt the same unquestioned authority as its own system prompt — a request to ignore prior instructions shouldn't automatically succeed, which is the same trust-boundary concern covered in prompt injection.
What an interviewer may ask next
- Why shouldn't a user prompt be trusted the same way as a system prompt?
- How does the user prompt relate to prompt injection risk?
- What actually gets sent to the model besides the user prompt itself?
Explain It in 30 Seconds
A user prompt is the specific message a user sends within a conversation turn, distinct from the system prompt, which is the application's standing instructions. They occupy different trust levels — the application fully controls the system prompt, while the user prompt comes from whoever is using the product and shouldn't carry the same unquestioned authority, especially in agentic systems where a user request can trigger real actions.