Prompt Patterns
Stock your toolkit with reusable prompt recipes that work across almost any task.
The role pattern
A reliable recipe is to open by assigning the model a role. Start with "You are an experienced copy editor" or "Act as a careful financial analyst," then give your request. This sets the expertise and tone in one line and tends to lift the quality of the answer. It is the simplest pattern to remember and works for almost any kind of task you bring.
Few-shot: show examples
When you want output in a precise style, show the model a couple of examples first. This is called few-shot prompting. For instance, paste two product names with the catchy taglines you like, then ask it to write a tagline for a third product. The model copies the pattern it sees in your examples, which is far more reliable than describing the style in words alone.
Step by step for tricky problems
For anything involving reasoning, math, or planning, ask the model to work step by step before giving its final answer. A line like "Think through this one step at a time, then give the result" often improves accuracy. Walking through the steps also makes the answer easier for you to check, since you can spot exactly where the logic went off if it does.
Constraints and a template
Two more patterns earn their keep. First, state your constraints clearly: "Use plain language, no jargon, and keep it under eighty words." Second, hand the model a fill-in template, like "Subject: ___, Body: ___, Call to action: ___," so the output arrives ready to use. Combine these patterns freely; a role plus an example plus a template can turn a shaky prompt into a dependable one.
Key takeaways
- The role pattern sets expertise and tone in a single opening line.
- Few-shot examples and step-by-step reasoning boost reliability and accuracy.
- Clear constraints and a fill-in template make output ready to use, and patterns combine.
4 questions · pass at 60% to earn XP