Zero-Shot Prompting
Zero-shot prompting asks a model to perform a task from instructions alone, with no examples — the default mode for capable modern LLMs.
Zero-shot prompting is instructing a model to do a task with no demonstrations — just the description: "Classify this ticket as billing, bug, or feature-request. Respond with the label only."
The term comes from the ML literature (performing a task with zero training examples), and its practicality is a product of instruction tuning: models are explicitly trained to follow natural-language task descriptions, so clear instructions alone now cover most well-defined work. That makes zero-shot the sensible starting point — cheapest in tokens, easiest to maintain, no example set to curate or go stale.
Its limits define when to escalate: when output must match a pattern easier shown than told, add few-shot examples; when the task needs visible multi-step reasoning on a non-reasoning model, add chain-of-thought; when code consumes the output, enforce structure rather than describing it. The escalation path — and when each step actually pays — is the subject of Few-Shot vs Chain-of-Thought vs Structured Prompting.
Frequently asked questions
- Is zero-shot worse than few-shot?
- Not inherently — modern instruction-tuned models handle most well-specified tasks zero-shot, and examples cost tokens on every call. Few-shot earns its cost when outputs must match a pattern that's hard to verbalize (formats, style, fuzzy boundaries). The practical rule: start zero-shot with sharp instructions; add examples to fix the specific failures you observe.
- What makes a zero-shot prompt work well?
- Specification quality. State the role, the task, the constraints, and the exact output format; bound the edge cases ('if no date is present, return null'). Most zero-shot failures are underspecification wearing a model-quality costume — the model guessed because the prompt left the decision open.
Related
- Few-Shot PromptingFew-shot prompting includes worked examples in the prompt so the model learns the task's pattern from demonstrations instead of instructions alone.
- Few-Shot vs Chain-of-Thought vs Structured Prompting: What to Use When (2026)When to reach for few-shot examples, chain-of-thought reasoning, or structured/output-constrained prompting — a 2026 decision guide to the core techniques.
- System PromptThe system prompt is the standing instruction layer an LLM receives before user input — defining its role, rules, tools, and tone for the whole conversation.
- Prompt Patterns for Coding AgentsPractical prompting patterns: chaining, few-shot, context management, tool use, and output structuring.