Language models respond to what you give them. A vague prompt produces a vague answer, while a prompt with context, constraints, and an example produces something usable. The skill is not tricking the model. It is writing a brief the way you would brief a capable colleague: background, task, limits, and a sample of the desired result.
Parts of a strong prompt
- Context: the facts the model needs, pasted in, since it cannot know your data or situation otherwise.
- Task: one clear instruction, stated as a verb, such as summarise, compare, or draft.
- Constraints: length, format, tone, and what to leave out, because unconstrained output wanders.
- Example: one input and output pair showing the shape you want, which guides the model better than adjectives.
When an answer misses, change one part at a time. Changing everything at once teaches you nothing about what fixed it.