The five components
Nearly every strong prompt contains the same five parts, in this order:
- Role — the expertise to apply. "You are a technical editor for a developer audience."
- Context — the situation, audience and prior facts. This is the part people skip and the part that matters most.
- Task — the single thing to do. One prompt, one job.
- Constraints — length, tone, reading level, what to avoid, what must be included.
- Format — the exact output shape: headings, table columns, JSON keys, bullet count.
A worked upgrade
Weak: "Write about our new pricing."
Strong: "You are a SaaS product marketer. We are moving from three plans to two and raising the entry price from $9 to $12 for new customers only; existing customers keep their price for twelve months. Write the in-app announcement for existing customers. Constraints: under 120 words, reassuring but not apologetic, no marketing adjectives, lead with the fact that their price is unchanged. Format: one short paragraph, then three bullets, then a one-line link CTA."
The second prompt is longer to write and roughly ten times more likely to produce something publishable, because every decision the model would otherwise guess at has been made by the person who actually knows the answer.
Which component to add
Diagnose by symptom:
| Symptom | Missing component |
|---|---|
| Correct but generic | Context |
| Right content, wrong shape | Format |
| Too long, too flowery | Constraints |
| Shallow or naive treatment | Role |
| Answers several things badly | Task (split it) |
Templates beat inspiration
Once a prompt works, save it with placeholders and reuse it. That is precisely what the tools on this platform are: a stored system prompt, a role, a quality bar and an output format, with a form for the two or three variables that actually change. The Prompt Library applies the same idea to free-form chat.
Practice
Take your last disappointing AI result. Add exactly one missing component from the table above and rerun. Repeat. You will usually find that two additions — context and format — account for most of the gap.