Few-shot prompting includes a handful of worked examples in the prompt so the model can infer the pattern. Zero-shot prompting provides none. Examples improve consistency and output format, but every example adds input tokens - so measure the quality gain against the added cost.
Why it matters
Examples communicate format and intent faster than prose instructions, often replacing several paragraphs of rules. The trade-off is input tokens: every example is billed on every request.
How it works
Few-shot prompting places input/output pairs in the prompt and lets the model infer the pattern, then applies it to the real input. Zero-shot provides no examples and relies on instructions alone.