Definition
A prompt is the text input provided to a language model that instructs it what to do or respond to. It can range from a simple question to complex multi-part instructions including context, examples, and formatting requirements. The art of crafting effective prompts—prompt engineering—is crucial for getting high-quality outputs from language models.
Why it matters
Prompts are the primary interface between users and language models:
- Output quality — well-crafted prompts dramatically improve response accuracy and relevance
- Task definition — prompts tell the model what task to perform (summarize, translate, analyze)
- Behavior control — prompts can set tone, format, length, and constraints
- Zero-shot learning — good prompts enable models to perform tasks without fine-tuning
The same model can produce vastly different outputs depending on how it’s prompted.
How it works
┌────────────────────────────────────────────────────────────┐
│ PROMPT STRUCTURE │
├────────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ SYSTEM PROMPT (sets behavior/persona) │ │
│ │ "You are a helpful tax advisor..." │ │
│ └─────────────────────────────────────────────────────┘ │
│ │ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ CONTEXT (retrieved documents, prior conversation) │ │
│ │ "Based on the following tax regulations..." │ │
│ └─────────────────────────────────────────────────────┘ │
│ │ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ USER PROMPT (the actual question/task) │ │
│ │ "Explain the deduction rules for home offices" │ │
│ └─────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ MODEL RESPONSE │
└────────────────────────────────────────────────────────────┘Key prompt components:
- System prompt — persistent instructions that define model behavior
- Context — background information or retrieved documents
- Examples — demonstrations of desired input/output pairs (few-shot)
- User query — the specific question or task
- Output format — specification of how to structure the response
Common questions
Q: What is prompt engineering?
A: Prompt engineering is the practice of designing and optimizing prompts to get better results from language models. It includes techniques like chain-of-thought, few-shot examples, and structured formatting.
Q: What’s the difference between zero-shot and few-shot prompting?
A: Zero-shot gives no examples—just instructions. Few-shot includes examples of the desired input/output pattern. Few-shot typically improves accuracy for complex tasks.
Q: How long should a prompt be?
A: As long as needed for clarity, but within context limits. More context isn’t always better—focused, well-structured prompts often outperform verbose ones.
Q: Can prompts be “jailbroken”?
A: Adversarial users sometimes craft prompts to bypass safety guidelines. This is why production systems need robust prompt injection defenses and content filtering.
Related terms
- LLM — models that respond to prompts
- System Prompt — persistent configuration prompt
- Context Window — limits total prompt size
- Chain-of-Thought — prompting technique for reasoning
References
Brown et al. (2020), “Language Models are Few-Shot Learners”, NeurIPS. [25,000+ citations]
Wei et al. (2022), “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models”, NeurIPS. [5,000+ citations]
Liu et al. (2023), “Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in NLP”, ACM Computing Surveys. [3,000+ citations]
Reynolds & McDonell (2021), “Prompt Programming for Large Language Models”, arXiv. [500+ citations]