AI Prompts That Actually Work
The internet is full of ‘100 magic prompts’ lists, and almost none of them survive contact with real work. That’s because good prompting was never about magic words — it’s about task specification. This card shows you the anatomy of a prompt that works, why the bad ones fail, and how to iterate purposefully instead of re-rolling and hoping.
Why this matters
Prompt libraries age in weeks and rarely fit your exact task, your audience or your organisation’s tone. What doesn’t age is the underlying skill: describing work precisely enough that a capable assistant — human or artificial — can contribute usefully. Learn the anatomy once and you can construct the right prompt for any task, which beats owning a thousand prompts for tasks you don’t have.
Why prompts fail
- No context — the tool knows nothing about your project, audience or history unless told
- No role for judgement — asking AI to decide things only you can decide
- No output shape — so you get an essay when you needed a table
- One-shot thinking — treating the first response as the final answer instead of a first pass
The anatomy of a working prompt
A working prompt is a five-part brief in a single message: goal, context, audience, constraints, output shape. Here’s one from real workplace life:
‘I need a status update on our records migration project for the operations manager. Context: we’re two weeks behind because the legacy export ran slower than planned; the vendor has added capacity; the final deadline is unchanged. Audience: a busy manager who wants honesty without drama. Constraints: under 250 words, plain English, no blame, don’t promise a catch-up date I haven’t given you. Output: three short sections — where we are, what changed, what happens next.’
Every sentence in that prompt is doing a job. Nothing in it is a magic word.
Iterating with purpose
- Read the first draft and name the gap: wrong content, wrong tone, wrong shape, or wrong emphasis
- Fix the corresponding part of the brief — content gaps mean missing context; tone gaps mean unspecified audience
- Ask for the change directly: ‘keep everything, but make the risks section more direct’
- Stop at good enough to finish yourself — the last 20 per cent is your judgement anyway, and polishing prompts past that point wastes the time you saved
Putting it into practice
- Take a prompt that disappointed you recently and rewrite it with all five parts
- Build reusable briefs for your two most repetitive tasks
- Practise one purposeful iteration instead of three hopeful re-rolls
- Strip ‘magic words’ you’ve cargo-culted; replace them with actual context
- Share your best working prompt with a colleague — theirs will teach you something too
Key takeaways
- Prompting is task specification, not incantation
- Working prompts contain goal, context, audience, constraints and output shape
- Diagnose weak results by naming the gap, then fix that part of the brief
- Iterate purposefully; stop at good-enough-to-finish-yourself
- The skill outlives every tool and every prompt list