Zero-Shot Prompting

Zero-shot prompting means giving a language model a task description and the input, with no example demonstrations. The model relies entirely on what it learned during training and on how clearly the instruction is written.

Why it matters

It is the simplest and cheapest way to use a model: short prompts, fewer tokens and no need to curate examples. Modern instruction-tuned models handle many common tasks, such as summarising, translating or labelling sentiment, well zero-shot. When results are inconsistent in format or miss subtle conventions, moving to few-shot prompting or adding a stricter output specification is the usual next step.

Example

Extract the domain name from this sentence and reply with only the domain.

"Please check whether status.example.org is reachable."

Expected reply: status.example.org. The instruction states the task and output format, but shows no sample pairs.

Common mistakes

• Writing vague instructions such as "analyse this" and then blaming the model for an unfocused answer.
• Not specifying the output format, which makes responses hard to parse in code.
• Assuming zero-shot accuracy on a niche or unusual task without testing it on real samples.

Related

• Few-shot prompting
• Prompt
• Prompt engineering
• LLM

This page is general information written by our editorial team and updated when documentation changes. Servers and providers differ, so test changes safely first. See our editorial policy and disclaimer.