Skip to main content
Contact

AI glossary Models & Architecture

What is zero-shot learning and zero-shot prompting?

← All glossary terms
In short

A model's ability to perform a task it has not been shown examples of, guided only by an instruction or description of the task.

Explained

Zero-shot means a model handles a task without being given examples of that task, either in training or in the prompt. With large language models, zero-shot prompting is simply describing what you want, such as classify this email or summarise this contract, and letting the model do it. It works because large models learn broad patterns of language and tasks during pre-training. Zero-shot is the cheapest and fastest way to start, but for specialised formats or domains, adding a few examples (few-shot) or fine-tuning often improves accuracy and consistency.

Zero-shot, one-shot and few-shot

  • Zero-shot: only the instruction. Example: classify the sentiment of this review as positive, negative or neutral.
  • One-shot: the instruction plus one worked example.
  • Few-shot: the instruction plus several examples, which show the model the expected format and edge cases.

The GPT-3 paper made these terms common by showing that large models could do many tasks from instructions and a few examples in the prompt, without updating their weights.

When zero-shot works, and when it does not

Zero-shot tends to work well for general tasks the model has met in many forms during pre-training: summarising, translating, classifying into clear categories, extracting obvious fields or answering general questions.

It struggles when the task depends on conventions the model cannot guess: a company's own categories, a strict output format, specialised terminology or subtle judgement criteria. In those cases, clear instructions, a few good examples, retrieval of relevant context or fine-tuning usually help more than rewording the prompt.

Zero-shot reasoning

Zero-shot also applies to reasoning. Research has shown that simply asking a model to think step by step, with no worked examples, can improve its results on multi-step problems. Many current models are trained to reason before answering, which makes zero-shot use stronger for complex tasks.

Frequently asked questions

What is the difference between zero-shot and few-shot prompting?

Zero-shot prompting gives the model only the instruction. Few-shot prompting adds several examples of inputs and expected outputs, which usually helps with specific formats, categories or styles.

What is an example of zero-shot prompting?

Asking a model: 'Classify this customer email as complaint, question or request, and reply with the category only', without showing any classified emails first.

What is zero-shot classification?

It is assigning labels to text or images that the model was not explicitly trained to recognise, using only the label names or descriptions. Large language models and models such as CLIP are commonly used for it.

When should I use few-shot instead of zero-shot?

When zero-shot results are inconsistent, when the output must follow a precise format, or when categories and criteria are specific to your organisation. A handful of well-chosen examples often fixes those problems.

Is zero-shot the same as zero-shot chain of thought?

Not exactly. Zero-shot chain of thought is a specific technique: you add an instruction such as 'let's think step by step' so the model reasons before answering, still without giving examples.

Further reading

Sources

Updated: 2026-10-21

Want to see how this applies to your organisation?

Talk to an AI Expert