Few-Shot Prompting

Few-shot prompting is one of the most reliable ways to get consistent, format-perfect outputs from AI. Instead of describing what you want, you show it — giving the model 2–5 examples of exactly the input/output pattern you need.

The name comes from machine learning: "zero-shot" means no examples, "one-shot" means one example, "few-shot" means a small number of examples.

Why Examples Beat Instructions

When you describe what you want in words, the model interprets your description — and interpretation introduces variability. When you show an example, the model pattern-matches directly — which is more reliable.

Think of it like hiring a contractor. Saying "I want clean, modern design" is vague. Showing them three photos of exactly what you mean is precise.

Zero-shot (description only):
"Rewrite these emails to be more professional"

Few-shot (description + examples):
"Rewrite these emails to be more professional.

Example 1:
Informal: 'hey can we meet tmrw about the project?'
Professional: 'Could we schedule a meeting tomorrow to discuss the project?
Please let me know your availability.'

Example 2:
Informal: 'the deadline got pushed back no worries'
Professional: 'I wanted to inform you that the deadline has been extended.
No immediate action is required on your end.'

Now rewrite: 'got ur email, will check it out later'"

The few-shot version maintains consistent tone, formality level, and sentence structure — exactly because the examples anchored the pattern.

The Structure of a Few-Shot Prompt

[Task description]

[Input label]: [Example 1 input]
[Output label]: [Example 1 output]

[Input label]: [Example 2 input]
[Output label]: [Example 2 output]

[Input label]: [Example 3 input — optional but strengthens the pattern]
[Output label]: [Example 3 output]

[Input label]: [Your actual input]
[Output label]:

Leave the final output label empty — the model fills it in following the established pattern.

High-Value Use Cases

Consistent content generation:

Generate product descriptions in our brand voice.

Product: AirPods Pro
Description: Immerse yourself in pure sound. AirPods Pro's
Adaptive Transparency lets you tune out the world — or let it in.
Active Noise Cancellation that actually works.

Product: Apple Watch Series 9
Description: The watch that thinks. New Double Tap gesture.
Precision Finding for iPhone. Carbon neutral. Your health,
your life, on your wrist.

Product: [Your product name]
Description:

Data extraction and transformation:

Extract key information from job descriptions as JSON.

Job posting: "Senior Software Engineer at Stripe. 5+ years Python required.
Remote-friendly. $150K-$200K. Must have distributed systems experience."
Output: {"title": "Senior Software Engineer", "company": "Stripe",
"years_exp": 5, "salary_min": 150000, "salary_max": 200000,
"remote": true, "key_skills": ["Python", "distributed systems"]}

Job posting: "Marketing Manager at Shopify. 3+ years experience.
Toronto office required. $80K-$100K CAD. HubSpot knowledge preferred."
Output: {"title": "Marketing Manager", "company": "Shopify",
"years_exp": 3, "salary_min": 80000, "salary_max": 100000,
"remote": false, "key_skills": ["HubSpot"]}

Job posting: [Paste job description here]
Output:

Classification tasks:

Classify customer support tickets by urgency (1=Low, 2=Medium, 3=High, 4=Critical).

Ticket: "The font on the billing page looks slightly different from the rest of the site."
Urgency: 1

Ticket: "I've been charged twice for my subscription this month."
Urgency: 3

Ticket: "Our entire team can't log in. This is blocking a product launch in 2 hours."
Urgency: 4

Ticket: [Customer message]
Urgency:

How Many Examples to Use

Situation	Examples Needed
Simple, clear patterns	1–2
Nuanced tone/style matching	3–5
Complex structured output	2–3 with varied inputs
Highly consistent formatting	3–5

More than 5 examples rarely improves performance and wastes context window space.

Choosing Good Examples

The quality of your examples matters as much as the quantity. Good examples should:

Represent the range of inputs you expect (not just the easy cases)
Show the exact format you want in outputs
Cover edge cases if consistency there matters
Be clean and unambiguous — confusing examples confuse the model

Combining Few-Shot with Role-Based Prompting

The most powerful combination:

"You are a UX writer at Apple, known for copy that is clear,
human, and empowering — never corporate.

Example 1:
Feature: Face ID on iPhone
Copy: 'Your face is your password. Unlock everything instantly —
securely, privately, just for you.'

Example 2:
Feature: iCloud Backup
Copy: 'Everything you care about, safe in the cloud. Automatic.
Invisible. Always there.'

Feature: [Your feature]
Copy:"

The role establishes the expert voice; the examples establish the exact format and style.

Practice

Build a personal few-shot prompt library for your 3 most common AI tasks. For each task, write 3 good examples of your ideal output. Store these and reuse them — they're reusable assets that consistently produce high-quality results.