How to write better ChatGPT and Claude prompts
Text models usually perform best when they can distinguish the task from the background information. A strong prompt is not necessarily long; it is specific enough to reduce ambiguity.
The five-part framework
- Task: Say what you want the assistant to produce.
- Context: Provide the situation, audience, or source material.
- Constraints: Set limits on length, tone, scope, and assumptions.
- Format: Request a table, email, checklist, or another usable structure.
- Verification: Ask it to identify unknowns rather than inventing details.
Before and after
Basic: “Write a customer email.”
When not to over-engineer
Simple questions often need simple prompts. Add constraints only when they help the task. Always review factual claims and citations before publishing or sending the result. You can experiment with the Make Prompt AI text mode.
Writing useful prompts for text models
A useful prompt identifies the task, audience, constraints, input data and output format. Give the model enough information to distinguish verified facts from assumptions. Ask it to flag uncertainty instead of inventing missing details.
Try a before-and-after example
Basic: Write a customer support email. Improved: Draft a calm, 120-word email explaining a scheduled 20-minute maintenance window, including a subject line, expected impact and support contact placeholder. Do not promise zero downtime.