Few-shot prompting means giving the AI a handful of examples of what you want before asking it to do the task. Instead of describing the format in words, you show it: two or three input-and-output pairs, then your real input. The model infers the pattern and follows it. ("Zero-shot" is the opposite — no examples, just the instruction.)
It's the highest-leverage prompting trick most people never use, and it requires no technical skill — just paste examples.
Why it matters at your desk. When the shape of the output matters, examples beat description every time. A teacher who wants quiz questions in a specific style gets far better results by pasting two model questions than by describing them. A marketer who shows the AI three on-brand product blurbs gets a fourth that matches the voice — which is why tools like Jasper and Lavender lean on saved examples and templates under the hood. It's a lighter-weight alternative to fine-tuning: you steer with examples in the moment instead of retraining the model.
What to watch for: your examples are the instruction, so a sloppy or inconsistent example teaches sloppiness. Pick examples that are correct, consistent, and representative — and two strong ones usually beat five mediocre ones.