Glossary · Technique
Zero-Shot Prompting
Also known as: Instruction-only prompting
No examples — just the instruction. The baseline every prompt is implicitly trying to beat.
When to use it
- Simple, well-known tasks the model already does well.
- When you don't have good examples to provide.
- Quick chat-style use — most consumer prompts are zero-shot.
- When token budget is tight.
When not to use it
- Anywhere format consistency matters across many calls.
- Tasks with non-standard structure where the model defaults to a different one.
- When the failure rate is unacceptable — few-shot almost always wins for accuracy.
How it works
- 1You provide just the instruction (and optionally a role / format / length).
- 2The model relies entirely on its pre-training and instruction-tuning to figure out what you want.
- 3Modern instruction-tuned models (GPT-4, Claude 3+, Gemini 1.5+) are dramatically better at zero-shot than earlier models.
Example
Lazy prompt
Translate this to French: hello
Using the technique
Translate the following English text to French. Return only the translation, no commentary. Text: Hello, can we reschedule tomorrow's meeting to next week?
Common pitfalls
- Output format drifts across calls — you can't rely on consistency.
- Edge cases fall through; the model defaults to its most common pattern.
- When zero-shot fails, the response is often confidently wrong.
Where this came from
Implicit in all LLM prompting; named explicitly in contrast to few-shot in the GPT-3 paper.
Related techniques
Few-Shot Prompting
Show the model 2–5 examples of input → desired output before asking. Consistently the highest-ROI prompt move in real-world AI work.
Chain-of-Thought (CoT) Prompting
Force the model to think step-by-step before answering. Dramatically improves accuracy on multi-step problems.
Persona / Role Prompting
Tell the model who it is. "Act as a senior X" changes accuracy, vocabulary, and tone more than any other single trick.