Zero-Shot, Single-Shot, and Few-Shot Prompts

There's no single right way to write AI prompts. Different prompts offer different paths to success.

In this post, we'll explore three different kinds of prompts - from simple to complex - and when to use each one.




Zero-shot Prompt

A zero-shot prompt is a single request where you provide the AI with instructions but no examples of the desired output. This type of prompt requires the AI to perform the task using only its prior knowledge. Quick facts, simple text and data transformations work well with a zero-shot prompt.

While zero-shot prompts are simpler and use fewer tokens, you have less control over the format, tone, and accuracy of the response.

Here's an example of a zero-shot prompt:

Question: What U.S. state has the most national parks?
Answer with **only** the state name.

Single-shot Prompt

When prompts move beyond simple requests, you may need to provide an example of the input and output. A prompt that includes a single example of what you expect is known as a single-shot (one-shot) prompt.

In a single-shot prompt, you include the input/output to demonstrate to the AI the tone, format, or reasoning that you expect, followed by the task you want to solve, which gives you more control over the response compared to zero-shot prompts.

Let's look at an example. Here, we're asking for the most iconic feature of a U.S. National Park. We've included one example of what we expect, followed by our question:

You are a U.S. National Park guide. For each park, reply with exactly one short phrase naming its most iconic natural or cultural feature.

Example
Arches National Park → Delicate Arch

Now answer:
Zion National Park →

While a single-shot prompt gives you more control over the tone or formatting, there are cases when a single example may not be enough to capture patterns or edge cases.

Few-shot Prompt

For complex prompts that require consistent formatting or nuanced responses, we need to include multiple examples. This type of prompt is known as a few-shot prompt.

If we take our single-shot prompt and add more examples of what we expect, we now have a few-shot prompt. By providing more examples, we've made our prompt more consistent, and we can be more confident in the results we get back.

You are a U.S. National Park guide. For each park, reply with exactly one short phrase naming its most iconic natural or cultural feature.

Arches National Park → Delicate Arch
Grand Canyon National Park → South-Rim Canyon Vistas
Yellowstone National Park → Old Faithful Geyser
Everglades National Park → Sawgrass Marsh

Now answer:
Zion National Park →

Because few-shot prompts require more examples, they have a higher token cost and may also experience increased latency. However, this trade-off is often worthwhile for tasks that require precise formatting or complex reasoning.

Choosing the Right Approach

Knowing when to use each type of prompt will help you become more effective with AI. If you're unsure where to start, begin with a zero-shot prompt and then add examples if the response accuracy drifts or if you need more control over the output format.

Happy prompting!