What is zero-shot prompting?

Zero-shot prompting means asking a model to do a task without giving it any examples — you just describe what you want and let the model rely on the knowledge it picked up during training. "Translate this sentence to French" or "Is this review positive or negative?" with no sample answers attached is zero-shot. The model works it out from scratch, using patterns it absorbed from the huge amount of text it was trained on.

How is it different from few-shot prompting?

The difference is examples. Zero-shot gives none — instruction only. Few-shot prompting includes a handful of worked examples in the prompt — a few reviews already labeled positive or negative, say — so the model can copy the pattern before it handles your real input.

Few-shot is basically showing your work before asking the model to show its own. There is also one-shot, which is exactly what it sounds like: a single example, a middle ground between the two.

• Zero-shot — instruction only, no examples. Shortest prompt, leans entirely on pretraining.
• Few-shot — instruction plus a few input-output examples that demonstrate the exact format and style you expect.

When is zero-shot enough?

For common, well-understood tasks, zero-shot usually just works. Summarizing an article, answering a general question, rewriting a paragraph, classifying obvious sentiment — modern models have seen so much of this during training that examples add little. If the instruction is clear and the task is familiar, start zero-shot.

It is faster to write and uses fewer tokens.

When should you add examples?

Reach for few-shot when zero-shot gives inconsistent results. Add examples when you need a very specific output format, when the task is unusual or domain-specific, when the model keeps misreading an edge case, or when tone and structure matter and describing them in words is not landing. A couple of good examples often fixes what a longer instruction cannot.

The practical habit: try zero-shot first, and only spend tokens on examples once you have seen where the plain instruction falls short.