You asked for a sentiment classification. The model gave you a three-sentence essay. Why?
In Part 3 of our Prompting Techniques series, we cover the techniques that fix this — few-shot prompting, role assignment, positive instruction framing, and chain of thought prompting. We break down zero-shot vs few-shot with a real example that shows exactly how one inserted message completely changes the model's output format. Then we walk through how assigning a role activates a whole cluster of reasoning patterns, and why chain of thought prompting gets you better answers by guiding the model step by step.
This is the final part of the 3-part Prompting Techniques series. If you missed Parts 1 and 2, check the playlist — each one builds on the last.