Few-shot Prompting

In plain English

Few-shot means showing the AI a few finished examples in your prompt so it copies the pattern.

Analogy

Like showing a new hire two or three completed samples before asking them to do the next one.

Why it matters

Adding a few examples sharply improves accuracy and formatting consistency for tricky extraction or classification steps.

In practice

Showing three messages already labeled Sales/Support/Spam, then asking the model to label a fourth.

Related terms

Sources

• Anthropic prompting best practices: Use examples effectively
• Brown et al.: Language Models are Few-Shot Learners
• Gemini API: Prompt design strategies (zero-shot vs few-shot)