Module 2 of 5 · Beginner · 8 min read
Context and examples
The two strongest levers in prompting, and how to use them without overdoing it.
- 1Missing context is the most common cause of generic output.
- 2Two to five varied examples teach format and tone better than rules.
- 3Explain the reason behind important constraints.
01The context you forget to give
You know your audience, your company, your previous attempts and your constraints. The model knows none of it unless you say so. Before sending a prompt, ask: what would a smart new colleague need to know to do this well?
Paste the relevant material (the brief, the data, the draft) directly rather than describing it. Put long documents first and your request last.
02Few-shot examples
Examples show the model the pattern: length, structure, voice and judgement. For classification and extraction they are often the difference between 70% and 95% accuracy.
Models copy examples closely, so vary them. If all your examples start with a question, every output will too. Include an edge case if you care about how it is handled.
03Rules with reasons
"Never use ellipses" will be followed rigidly. "This text will be read by a speech engine that cannot pronounce ellipses, so never use them" lets the model avoid other things a speech engine would trip on, too.
A reason helps the model generalise to cases you did not anticipate.
Finished reading?
Mark it complete to track your progress, then test yourself with the Prompting Fundamentals quiz.