Definition
Zero shot prompting means giving an AI model a task with no examples attached. Few shot prompting means including one or more examples inside the prompt to show the model the pattern you want it to follow.
The distinction matters because examples cost tokens and take time to write, but they also steer the output in ways that instructions alone sometimes cannot. Neither approach is universally better. The right choice depends on how predictable the task is and how much the default output format differs from what you actually need.
You can read a deeper treatment of zero shot on the What Is Zero Shot Prompting? glossary page.
A worked example
Say you need the model to classify customer emails as complaints, questions, or compliments.
Zero shot: You describe the task and trust the model to figure it out.
Classify this email as "complaint", "question", or "compliment". Reply with one word only.
Email: "I've been waiting three weeks for my order and nobody has responded to my tickets."Few shot: You add examples before the live input.
Classify each email as "complaint", "question", or "compliment". Reply with one word only.
Email: "Your support team called me back within the hour." → compliment
Email: "Do you ship to Canada?" → question
Email: "I've been waiting three weeks for my order and nobody has responded to my tickets." →The second prompt is more reliable on edge cases because the model has seen what your labels actually mean in practice, not just what they mean in theory.
When does each approach work best?
Few shot prompting earns its extra setup when the output format is strict (structured JSON, a specific table layout, a branded tone) or when zero shot keeps drifting in the wrong direction. Zero shot is usually enough for open-ended tasks: summaries, brainstorming, simple Q&A.
A rough guide:
| Situation | Better choice |
|---|---|
| Novel task, flexible output | Zero shot |
| Strict format required | Few shot |
| Consistent brand voice | Few shot |
| Quick, one-off question | Zero shot |
Related terms
Zero shot prompting is the baseline: a prompt with no examples, relying entirely on the model's existing knowledge.
Few shot prompting adds two to five examples inside the prompt to show the model the format or tone you want.
System prompt sets persistent instructions before the conversation begins, and often works alongside either approach.
Chain-of-thought prompting asks the model to show its reasoning step by step, and pairs naturally with few shot when accuracy matters most.
Frequently asked questions
Is few shot prompting always better than zero shot?
Few shot prompting produces more consistent output when format matters, but it takes longer to write and adds tokens to every request. For simple, open-ended tasks, zero shot is often the right call.
How many examples count as "few shot"?
Two to five examples is the typical range. Beyond that, you are closer to fine-tuning territory than prompting.
Can I mix both approaches in one prompt?
Yes. A system prompt can set the task, and one or two inline examples can anchor the format.
Read the full guide
The choice between zero shot and few shot prompting is one of the first decisions worth understanding when you're building prompts that need to work consistently across a team. The full guide on prompting techniques covers both approaches in depth, alongside chain-of-thought prompting, role prompting, and others.
If you want ready-made prompts that already apply these techniques, the Convergence prompt template library has options built for real team workflows.