Zero-shot: just ask
Zero-shot prompting means you ask the model to do a task with no examples, just an instruction. 'Classify this review as positive or negative' is zero-shot. Modern models are surprisingly good at zero-shot performance on common tasks because they've seen millions of similar instructions during training, so for well-known task types like summarizing, translating, or answering factual questions, zero-shot is often enough.
The catch is that zero-shot works best when the task is common and your desired output format is standard. If you want a very specific style, an unusual format, or a nuanced judgment call, plain instructions alone often leave too much room for the model to guess wrong, and you'll get inconsistent results across multiple runs.