Course → Module 5: Prompt Engineering
Session 3 of 10

Show, Do Not Describe

Few-shot prompting means including examples of the desired output in your prompt. Instead of describing what you want in abstract terms, you show it. "Here are three product descriptions in my style. Now write one for this product." The AI pattern-matches against your examples and produces output that resembles them.

This works better than description because examples are unambiguous. The word "conversational" means different things to different people. An example of a conversational paragraph means exactly one thing. The AI does not interpret an example. It imitates it.

Few-shot examples teach the AI through demonstration. Three well-chosen examples communicate more about your desired output than a page of written instructions. The AI learns patterns: sentence length, vocabulary choices, structural habits, and tonal shifts. It learns what you do, not what you say you do.

Zero-Shot, One-Shot, Few-Shot

The terminology is simple. Zero-shot means no examples. One-shot means one example. Few-shot means two or more examples (typically three to five). Each additional example gives the AI more data to pattern-match against, but with diminishing returns.

graph LR A["Zero-shot
No examples
Relies on instructions only"] --> B["One-shot
One example
Establishes basic pattern"] B --> C["Few-shot (3-5)
Multiple examples
Robust pattern matching"] C --> D["Many-shot (10+)
Extensive examples
Diminishing returns, higher cost"] style A fill:#222221,stroke:#c47a5a,color:#ede9e3 style B fill:#222221,stroke:#8a8478,color:#ede9e3 style C fill:#222221,stroke:#6b8f71,color:#ede9e3 style D fill:#222221,stroke:#c8a882,color:#ede9e3
Approach Examples Token Cost Output Consistency Best For
Zero-shot 0 Lowest Low Simple tasks, brainstorming
One-shot 1 Low Moderate Straightforward formatting tasks
Few-shot 3-5 Moderate High Production content, voice matching
Many-shot 10+ High Marginal improvement Highly specific or unusual formats

Choosing Good Examples

The quality of your examples matters more than the quantity. Three well-chosen examples outperform ten mediocre ones. A good example demonstrates the specific qualities you want the AI to replicate.

Selection criteria for examples:

What Few-Shot Examples Teach (and Fail to Teach)

Examples effectively teach: sentence length patterns, vocabulary choices, structural habits, formatting conventions, and tone. The AI picks these up reliably because they are surface-level patterns in the text.

Examples poorly teach: logical reasoning, factual accuracy, depth of analysis, and original insight. These require the AI to understand the content, not just the form. Few-shot examples shape how the AI writes, not what it knows.

This distinction is critical. You can make AI output sound like you through examples. You cannot make AI output think like you. That requires context, constraints, and human review.

Combining Few-Shot With System Prompts

Few-shot examples and system prompts are complementary. The system prompt provides explicit rules. The examples provide implicit patterns. Together, they cover more ground than either alone.

A production prompt that uses both might look like this: a system prompt defining voice rules and constraints, followed by three examples of published content in the desired style, followed by the specific task. The AI reads the rules, absorbs the patterns from the examples, and generates output that reflects both.

Further Reading

Assignment

Collect three pieces of your own best writing (or writing you admire in a style you want to replicate). Use them as few-shot examples in a prompt asking the AI to produce similar content on a new topic. Then generate the same content without examples. Compare the two outputs. Write a brief analysis: what did the examples teach the AI? What did they fail to convey?