Prompt
The instructions and input you hand a model for one turn. Anything specific to your situation exists only if you put it in there.
See it
What it is
A prompt is the instructions and input you supply for one turn. The whole payload the model actually processes is bigger than that, and it is worth a separate name: the request context, which bundles system instructions, chat history, retrieved documents, pasted files, tool definitions and your prompt into one stream of tokens. The model is not blank outside that stream (plenty of general knowledge is baked into its weights), but anything specific to you, your docs, your data, this morning's ticket, only exists if you put it in.
Prompts that work share a shape: role (who the model is acting as), task (one imperative sentence), context (the raw material, clearly delimited), constraints (format, length, what to avoid), and examples for anything with a house style. Put the instructions before long context and restate the output format as the last line, because middles get skimmed.
Gotcha: 'be creative' and 'be concise' are wishes, not instructions. Models act on checkable specifics like 'under 120 words', 'no bullet lists', 'return only the JSON object'. And negative instructions leak: telling a model never to mention pirates puts the word pirates in its context. Say what you want instead of what you fear.
Ask AI for it
Rewrite my request below into a production prompt. Structure it as: a one-line role, a single imperative task sentence, the source material wrapped in labeled XML tags, an explicit output format naming every field, and 3 to 5 hard constraints stated positively. Add two worked input/output examples. Keep the instructions above the source material and repeat the output format as the final line. Return only the finished prompt, nothing else.