Few-Shot Mastery

Teach Claude any pattern with examples — from classification to structured extraction, with production code

What Is Few-Shot Prompting?

Few-shot prompting means giving Claude a few examples of the input-output pattern you want, then letting it generalize to new inputs. It is teaching by showing rather than explaining — and it works remarkably well because Claude can infer complex patterns from just 2-3 examples.

This is distinct from zero-shot prompting (no examples, just instructions) and one-shot prompting (a single example). For most tasks, 3-5 examples hits the sweet spot: enough to disambiguate the pattern without wasting context tokens.

Key insight: Few-shot examples are often more powerful than written instructions. Saying "classify sentiment as positive, negative, or neutral" is ambiguous — does "The acting was great but the plot was terrible" count as positive, negative, or neutral? An example resolves this ambiguity instantly.

Example 1: Input -> OutputClaude observes the first pattern

Example 2: Input -> OutputPattern recognition strengthens

Example 3: Input -> OutputClaude deeply understands the pattern

New Input -> Claude generates correct output!Pattern is applied to novel inputs

Zero-Shot vs. Few-Shot — Side by Side

Here is the same task done both ways. Notice how few-shot produces more consistent, predictable output:

Zero-shot (instructions only)

Classify this review's sentiment
as positive, negative, or neutral.

Review: "Decent food but the
service was painfully slow."

# Claude might say:
# "Negative" or "Mixed" or
# "The sentiment is primarily
#  negative with a positive
#  element..." (verbose)

Few-shot (with examples)

Review: "Loved every minute!"
Sentiment: Positive

Review: "Worst meal I've ever had."
Sentiment: Negative

Review: "It was fine, nothing special."
Sentiment: Neutral

Review: "Decent food but the
service was painfully slow."
Sentiment: Negative
# Consistent one-word answer
# matching the example format

The few-shot version produces exactly the format you showed — a single word. The zero-shot version might give a paragraph of analysis. Few-shot teaches both the logic and the format.

Few-Shot in the API

In the Claude API, few-shot examples go in the messages array as alternating user/assistant turns. Claude sees them as a conversation history and continues the pattern:

Python — few-shot sentiment classifier

import anthropic

client = anthropic.Anthropic()

def classify_sentiment(review: str) -> str:
    """Classify a review as Positive, Negative, or Neutral."""
    response = client.messages.create(
        model="claude-haiku-4-5-20251001",  # haiku for classification
        max_tokens=10,       # we only need one word
        temperature=0,       # deterministic
        system="Classify the sentiment of each review as exactly one word: Positive, Negative, or Neutral.",
        messages=[
            # Few-shot examples
            {"role": "user", "content": "The movie was absolutely fantastic!"},
            {"role": "assistant", "content": "Positive"},
            {"role": "user", "content": "I wasted two hours on this terrible film."},
            {"role": "assistant", "content": "Negative"},
            {"role": "user", "content": "It was okay, nothing special."},
            {"role": "assistant", "content": "Neutral"},
            # The real input
            {"role": "user", "content": review},
        ]
    )
    return response.content[0].text.strip()

# Use it
print(classify_sentiment("Great acting but terrible plot"))  # → Negative
print(classify_sentiment("A masterpiece of modern cinema"))  # → Positive
print(classify_sentiment("Meh"))                             # → Neutral

Notice: We use Haiku for classification — it is 4x cheaper than Sonnet and plenty smart for this task. We set max_tokens=10 because we only need one word. We set temperature=0 for deterministic output. These small optimizations add up at scale.

🔒

This lesson is for Pro members

Unlock all 520+ lessons across 52 courses with Academy Pro.

Go Pro — $49/mo ← Back to course

Already a member? Sign in to access your lessons.