Prompt Writing Guide

1. Start With a Single Outcome

Write one sentence for the exact outcome you want. If the task asks for multiple outcomes at once, split it into separate prompts.

2. Define the Role and Scope

Tell the model who it should act as and what is in scope. This prevents broad, generic responses.

Example: 'Act as a product marketer for a B2B SaaS onboarding flow. Focus only on activation emails.'

3. Add Context That Changes Decisions

Only include context that influences output quality: audience, constraints, source material, and decision criteria.

4. Set Hard Constraints

Hard constraints reduce drift and make evaluation easier.

• Length limits (for example: 120-160 words).
• Required sections (for example: summary, risk, recommendation).
• Forbidden content (for example: no legal claims, no invented metrics).

5. Specify Output Format

Ask for an explicit structure (bullets, table, JSON, markdown sections). Format control is one of the fastest ways to improve usability.

6. Add Evaluation Criteria

Tell the model how quality will be judged.

Example: 'Prioritize factual clarity, concise language, and clear next steps.'

7. Use a Two-Pass Workflow

Pass 1 generates a draft. Pass 2 critiques that draft against your criteria and rewrites weak sections.

8. Test With Edge Cases

Run the same prompt on at least three different inputs: ideal, minimal, and ambiguous. Keep the version that remains stable across all three.

9. Save Reusable Prompt Skeletons

Store reusable prompt structures by task type, then fill placeholders for each project.

10. Know the Limitation

Even strong prompts cannot replace source quality. If source facts are incomplete or wrong, output quality will still suffer.

If you want a quick sanity check, you can test your draft in this site's Prompt Quality Check tool before sending it to your model.