There's a belief that getting good AI results requires long, elaborate prompts: assign it a persona, provide paragraphs of context, add a list of rules. Sometimes that helps. More often, the result you want was available with one clear sentence, and the extra padding actively got in the way. I've written about this before in a simpler prompt framework, but it's worth revisiting because the instinct to over-explain never really goes away.
Why padding hurts
Every sentence in your prompt is something the AI tries to honor. Vague sentences ("make it sound professional but friendly, but not too casual") give it contradictory jobs. The model doesn't push back like an employee would, it just quietly averages your instructions into mush.
What a clear prompt contains
Three things, stated plainly:
- The task: what to produce. "Write a reply to this review."
- The constraints: the rules that matter. "Two sentences. Mention the refund. No exclamation points."
- The material: the facts it can't know. "We were late because a job ran over. Customer is a regular."
That's usually it. If a result misses, don't add padding, look at the output, find the gap, and state the missing constraint. Iteration beats incantation. It's the same instinct behind catching AI hallucinations: check the output against reality instead of trusting it because the prose sounds confident.
The golf test
A useful exercise: take a prompt that works and cut words until it stops working. You'll be surprised how much survives. What's left is the actual instruction, and seeing it teaches you more about prompting than any list of "magic phrases."
It works on people too
"Can you get me the March numbers by Friday, totals by service line" outperforms a paragraph of context in the same way. Clear asks get clear results. AI just gives you a place to practice where nobody gets annoyed. If you're handing this kind of work to AI regularly, that's exactly what an AI concierge is for: someone who already knows how to ask.