Zero-Shot Prompting

Simple Definition

Zero-shot prompting means giving an AI a task with no examples, just an instruction. The model uses only what it learned during training to figure out what you want and how to respond.

“Zero-shot” means zero examples. Most everyday AI use is zero-shot: you type a question or request and the AI responds without you providing any sample answers first.

Example

Zero-shot prompt:

Summarize the following article in 3 bullet points: [article text]

No examples are provided. The model already knows what “summarize” means and what bullet points look like from its training.

When Zero-Shot Works Well

Zero-shot works well for:

• Common, well-understood tasks (summarizing, translating, explaining)
• Tasks where quality doesn’t need to exactly match a specific format
• Quick, one-off requests where adding examples would take too long

When to Switch to Few-Shot

Use few-shot prompting when:

• You need a very specific output format
• The model’s zero-shot attempts keep missing the mark
• The task is unusual or highly specialized
• Consistency across many outputs is important

The Spectrum of Shots

Modern LLMs are surprisingly capable at zero-shot tasks, which is what makes them so useful without any setup.

Related Terms

• Few-Shot Prompting, providing examples to guide the model
• Prompt Engineering, the skill of writing effective prompts
• Chain-of-Thought, a technique for complex reasoning tasks
• LLM, models capable of zero-shot task completion