Prompt

A prompt is what you send to an LLM or image model before it generates output. Without a clear prompt, the model guesses intent; with structure (roles, examples, JSON schema), results become more reliable.

System-level prompts are set by the product owner, while user prompts are what you type in the input box. For deeper practice, see prompt engineering.


Key characteristics

  • Is the instruction or context you send to the model before it responds.
  • Can range from a simple question to a full package of roles, rules, examples, and attached context.
  • Directly affects tone, precision, format, and how much the model needs to guess.