Prompt
In one sentence A prompt is everything you send to a language model — the instruction, the context, the examples — and its quality largely decides the answer's quality.
Updated
A prompt is the full input text a language model receives — your question, plus any instructions, context and examples wrapped around it.
The best analogy is briefing a skilled freelancer who knows nothing about you. "Design a poster" gets you something. "Design an A3 poster for a college tech fest, young audience, dark background, these three sponsor logos, tagline in Hinglish" gets you roughly what you wanted. The freelancer's skill was constant; the brief changed. A model is that freelancer at the extreme: enormously capable, zero context about you beyond what the prompt carries.
Anatomy of a working prompt, in the order the pieces usually appear:
role/instruction what to do, and any rules ("Reply in JSON. Cite the source line.")
context the material to work from — the email thread, the data, the policy
examples 1-3 worked demonstrations (few-shot) when format matters
the actual ask the specific question or taskEverything competes for the same context-window, and the model weights what is present — it cannot honour constraints you left in your head. The craft of arranging these pieces is prompt engineering: specificity beats politeness, showing beats telling (few-shot), and asking for working-out (chain-of-thought) helps on reasoning tasks. In production, prompts live in code as templates with variables — treat them like source code: versioned, tested, reviewed.
Where to go next
- Full lesson: Prompt engineering
- Related terms: system-prompt, few-shot, context-window, chain-of-thought