A prompt is the text you give a model to condition its output, instructions, examples, context, and the question. Because an LLM only ever continues text, the prompt is the entire steering wheel: change the framing and you change the behaviour.
Useful moves: state the role and goal plainly, show a few examples (few-shot), ask the model to reason step by step, and supply grounding via retrieval. Everything in the prompt competes for space in the context window.