Examples and Few-Shot Prompting
Showing the AI a few examples of what you want is often the fastest way to get exactly that style or format.
Zero-Shot, One-Shot, Few-Shot
| Name | Meaning |
|---|---|
| Zero-shot | Only an instruction, no examples |
| One-shot | An instruction plus one example |
| Few-shot | An instruction plus several examples (usually 2 to 5) |
Zero-shot works for simple tasks. When you need a specific pattern, examples beat long explanations.
A Few-Shot Example: Labelling Reviews
Example
text
Label each customer review as Positive, Negative or Mixed.
Review: "Delivery was quick and the shoes fit perfectly."
Label: Positive
Review: "The colour was different from the picture."
Label: Negative
Review: "Good quality, but it took 10 days to arrive."
Label: Mixed
Review: "Excellent packaging and the seller replied fast."
Label:The model sees the pattern and continues it, answering "Positive".
Building Few-Shot Prompts in Code
When you have many items to label, you can build the prompt in a loop:
Example
python
examples = [
("Delivery was quick and the shoes fit perfectly.", "Positive"),
("The colour was different from the picture.", "Negative"),
("Good quality, but it took 10 days to arrive.", "Mixed"),
]
new_review = "Excellent packaging and the seller replied fast."
prompt = "Label each review as Positive, Negative or Mixed.\n\n"
for text, label in examples:
prompt += f'Review: "{text}"\nLabel: {label}\n\n'
prompt += f'Review: "{new_review}"\nLabel:'
print(prompt.count("Review:"), "reviews in the prompt")
print(prompt.splitlines()[-1])Output
4 reviews in the prompt Label:
Examples for Style
Examples also teach tone and style. Suppose you want product descriptions that sound like your brand:
Example
text
Write a product description in the same style as these examples.
Example 1: "Lawn suit in mint green. Soft, breathable, made for
Karachi summers. Ready to wear in 3 days."
Example 2: "Black cotton kurta. Simple cut, strong stitching,
looks sharp at work and at Friday prayers."
Now write one for: a maroon embroidered shawl for winter weddings.Choosing Good Examples
- Cover the variety: include each label or case you expect (positive, negative, mixed).
- Keep them consistent: same format in every example.
- Make them realistic: similar to the real inputs.
- Avoid hidden patterns: if all your positive examples are short, the model may think short means positive.
- Do not over-do it: a few good examples usually beat twenty average ones.
If the AI keeps getting the format wrong, stop explaining and show one perfect example. It usually fixes the problem.
Examples for Formats
Few-shot is perfect when you need a strict output shape, such as one line per item:
Example
text
Turn each sentence into "Name | City | Course".
Sentence: Ayesha from Hyderabad joined the Python course.
Output: Ayesha | Hyderabad | Python
Sentence: Bilal, who lives in Quetta, is studying graphic design.
Output:Exercise
Exercise
What is few-shot prompting?