Why this matters

Most disappointing answers come from a one-line ask. A prompt with role, task, context and format gives the model something to aim at.

The 20-minute experiment

1. 1Find a weak prompt you sent this week. Copy it into a note
2. 2Rewrite it in four labelled parts: Role (who it should be), Task (the one thing you want), Context (what it needs to know), Format (length and shape)
3. 3Run both versions in two separate chats so neither sees the other
4. 4Read both. Mark the sentences you'd have to fix before sending

Role: you are a [role]. Task: [one specific thing]. Context: [what you need to know]. Format: [length and shape]. Now respond.

• Old prompt found and saved
• New version has all four labelled parts
• Both run in separate fresh chats
• The four-part version needed fewer fixes before you could use it