Show, don't just tell

One of the most reliable ways to improve a model's output is to include examples of what you want.

This is called few-shot prompting. Instead of describing the format you need, you show it two or three times, and the model infers the pattern and continues it.

Ask for a list of book recommendations: vague. Ask for a list structured as "Title (Author, Year) — one-sentence reason I'd like it": the model will produce exactly that format, without being told a single rule about formatting.

This works because the model was trained on pattern completion. Three examples of a format constitute a very strong signal about what comes next. It's often more reliable than elaborate instructions, because instructions have to be interpreted; examples are unambiguous.

The flip side: examples can also lead the model astray. If your examples contain implicit assumptions or biases, the model will pick those up and extend them. What you show shapes what you get — not just the format, but the content and tone as well.