Few-Shot Learning Prompt Template: Zero-Shot vs. One-Shot vs. Few-Shot

These three terms describe how many worked examples a prompt includes before asking the model to complete a new case — and the right number depends on how much the task benefits from having a pattern to follow versus how much it just adds unnecessary length.

Zero-Shot: No Examples

Just a task description and the input to complete, with no demonstrations at all:

Classify the sentiment of this review as Positive, Negative, or Neutral.
Input: Setup was confusing but once it worked, I loved it.
Output:

This works fine for tasks the model already handles reliably from instructions alone — simple classification, straightforward rewrites, well-known formats. It's the fastest to write and the shortest prompt, at the cost of leaving more room for the model to interpret ambiguous instructions its own way.

One-Shot: A Single Example

One worked example added before the query — a reasonable starting point when a task's expected format isn't obvious from the instruction alone:

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

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

A single example establishes the expected output format clearly, but it can't demonstrate a full range of cases on its own — one positive example doesn't show how a negative or ambiguous case should be handled.

Few-Shot: Multiple Examples (Typically 2-5)

Enough examples to actually demonstrate the range of cases the task involves — one from each category, or examples spanning the difficulty the task presents:

Classify the sentiment of this 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:

This is genuinely "few-shot" in the sense the term is normally used — enough demonstrations to show a pattern across multiple cases, not just confirm a single format.

Choosing the Right Level

Start with zero-shot for straightforward tasks and only add examples if the output format or edge-case handling isn't coming out as expected. One example is often enough to fix a formatting issue; a genuinely ambiguous or multi-category task usually benefits from the fuller few-shot range. Adding examples has a real cost in prompt length, so it's worth adding only as many as the task actually needs to demonstrate its range.