Two of the most basic prompting styles are zero-shot and few-shot. The names sound technical but the idea is simple, and choosing between them is a everyday skill in working with models.
The difference
Zero-shot is asking the model to do a task with no examples — just an instruction: "Classify this review as positive or negative." Few-shot is showing a handful of examples first: "Review: 'Loved it' → Positive. Review: 'Waste of money' → Negative. Review: '...' →". The examples demonstrate the task and the exact format you want, rather than describing it.
Zero-shot tells the model what to do. Few-shot shows it. When words are ambiguous, showing beats telling.
When to use which
Few-shot shines when the task is hard to describe precisely, when you need a specific output format, or when there's nuance an example captures better than an instruction. Zero-shot is simpler, cheaper (no example tokens), and often enough for capable modern models that follow instructions well.
The shift
Early models needed few-shot examples for many tasks — they couldn't reliably follow a bare instruction. As models got better at instruction-following, zero-shot became viable for far more, and the trend moved toward clear instructions plus structured-output constraints instead of piling on examples. Few-shot remains valuable for format precision and edge cases, but it's no longer the default crutch it once was.
The practical rule
Start zero-shot with a clear instruction. If the output is inconsistent or the format drifts, add two or three well-chosen examples. Examples are especially powerful for pinning down format and demonstrating tricky cases — a couple of good ones often beats a paragraph of instructions.
