What is zero-shot prompting?
Zero-shot means giving a model a task with instructions only and no examples. The model relies on what it learned during training.
What is zero-shot prompting?
When you say "translate this into English" or "is this email a complaint or a question?", you give the model no examples; it understands the task from the description alone. That is zero-shot prompting. One of the early surprises of large language models was that they could do many tasks they were never specifically trained for, just from an explanation.
How it differs from few-shot
Few-shot prompting adds solved examples to the prompt. Zero-shot is shorter, cheaper and easier to prepare. For common, clear tasks such as summarizing, translation or general classification it is often enough. When the output format is strict or the task follows company specific rules, adding examples improves the result noticeably.
Getting better zero-shot results
A lot can be done without examples. Start by describing the task, the context and the output format clearly. For questions that need reasoning, asking the model to think step by step before answering can improve accuracy. A 2022 study showed that adding just "Let's think step by step" to a question sharply raised success on math problems. Newer reasoning models already do this step internally.
Which one when?
A practical rule: try zero-shot first. If results are inconsistent or the format breaks, move to few-shot with a handful of examples. If that is still not enough and the same narrow task repeats thousands of times, consider fine-tuning. Whichever you choose, a well written prompt is the foundation.