In plain terms
A prompt is a brief to a very capable colleague who knows nothing about your situation. Whatever you write is all they have: who the audience is, what a good result looks like, what to avoid. A one-line prompt gets a generic answer because the model fills every gap with the most average assumption.
Why it matters
The prompt is the main control anyone has over a model, and for most business uses the quality of the brief determines the quality of the output more than the choice of model does. Good prompts are also assets: they capture how a task should be done and can be versioned, tested and reused across a team.
Example
“Write a follow-up email” returns something bland. “Write a follow-up to a CFO who saw our demo on Tuesday and worried about integration time. Three short paragraphs, no jargon, one concrete next step. Here are my call notes: …” returns something close to sendable.
Most often confused with
Prompt vs. System Prompt
“Prompt” is the general word for input. In a chat application it has layers: the system prompt, written by the product's developer and usually invisible, and the user prompt, typed by the person. When users say “my prompt” they mean the second; the model sees both together.
Under the hood
Technically the prompt is the full token sequence sent to the model: system instructions, tool definitions, conversation history, retrieved documents and the latest message. Elements that reliably help: context and purpose, a specific task, the material to work on set apart by delimiters or tags, examples, the output format, and permission to express uncertainty. Order matters; long documents work best near the top and the question at the end. Prompts are sensitive to wording in ways that are hard to predict, so changes should be tested against a set of cases.