A capable language model can often perform a task it was never specifically trained for. Give it a few worked examples in the prompt, called few-shot, and it infers the pattern and continues it. Give it only an instruction with no examples at all, called zero-shot, and it may still manage from the description alone.
Examples usually help, especially for unusual formats, but they take up space and effort to craft. The practical skill is knowing when a clear instruction suffices and when a couple of examples will sharply improve the result, a judgement that saves both effort and frustration.