Few-Shot Prompt Builder: Show, Don't Just Tell the Model

Describing a task in words leaves room for a model to interpret it differently than intended. Showing a few worked examples first — input paired with the exact output wanted — removes most of that ambiguity. That's the entire idea behind a few-shot prompt, and building one by hand means carefully formatting each example consistently, which gets tedious past two or three.

What Goes Into the Prompt

• A task description — one line explaining what the model should do.
• Example pairs — each one an input and the exact output that input should produce.
• The actual query — the new input to complete, left without an output for the model to fill in.

Two Output Formats, Generated Together

The same examples produce two different formats automatically — a universal plain-text version, and an XML-tagged version optimized for Claude.

Plain Text (works with any model)

Classify the sentiment of each product review as Positive, Negative, or Neutral.
Input: The battery life is incredible, lasts two full days.
Output: Positive

Input: Arrived broken and customer service never replied.
Output: Negative

Input: It does what it says. Nothing more, nothing less.
Output: Neutral

Input: Setup was confusing but once it worked, I loved it.
Output:

XML-Tagged (Claude-optimized)

Classify the sentiment of each product review as Positive, Negative, or Neutral.

<example>
<input>The battery life is incredible, lasts two full days.</input>
<output>Positive</output>
</example>

<example>
<input>Arrived broken and customer service never replied.</input>
<output>Negative</output>
</example>

Now complete the following in the same way:
<input>Setup was confusing but once it worked, I loved it.</input>

Wrapping each demonstration in XML tags helps Claude specifically read them as reference examples rather than as additional tasks to respond to — a distinction that plain-text labeling alone can blur in longer prompts.