Prompt
The input you give an LLM — the question, instruction or context.
Often split in two: a system prompt that sets the role and rules ("you are a legal assistant that answers concisely") and a user prompt that's the actual question. How well you write them determines how good the answer is.
Prompt engineering is the craft of phrasing prompts for best results — often with techniques like "few-shot" (show examples), "chain-of-thought" (ask the model to think aloud) or structured output (ask for JSON).