Why most prompts underperform
When an AI gives you a bland or off-target answer, it's usually filling gaps you left open. It doesn't know who the answer is for, what you already know, or what "good" looks like to you, so it aims for the safe middle.
The fix isn't longer prompts for the sake of it. It's covering five things every time.
The five parts
Run through these before you hit enter:
- 1Role: who should the AI act as? ("You're a senior UK employment lawyer")
- 2Context: what's the situation, audience and background? Paste the relevant material.
- 3Task: one clear verb. Draft, compare, summarise, critique, plan.
- 4Format: bullets, a table, 150 words, a JSON object, an email with a subject line.
- 5Constraints: tone, things to avoid, reading level, what to do if unsure.
Before and after
A typical prompt, and the same request with the framework applied:
Before:
Write a LinkedIn post about our new feature.
After:
You're a B2B SaaS copywriter who hates buzzwords.
Context: We make invoicing software for UK plumbers. Today we launched
automatic payment reminders. Customers told us chasing late payments
is the worst part of their week. <notes>…paste release notes…</notes>
Task: Write a LinkedIn post announcing the feature.
Format: Under 120 words, a hook in the first line, 3 short paragraphs,
one call to action, no hashtags.
Constraints: Plain English, no exclamation marks, don't invent stats.Tip: Save the "after" version as a template with blanks. Next time you only fill in the context.
Make it ask you questions
If you can't be bothered to write all the context, flip it round: add "Before answering, ask me up to five questions you need answered to do this well." The model will tell you what's missing.
Checklist
Did I say who it's for? Did I paste the source material? Did I name the format? Did I say what to avoid? If all four are yes, you'll rarely need a second attempt.