If you've used ChatGPT or Claude and walked away thinking "this is mid," your prompts are probably the problem.
Most "prompt engineering" content gets weird. Long templates with role-playing instructions, ALL CAPS warnings, and "you are an expert in..." preambles. Skip all that. There's a 5-question template that works for 90% of business writing tasks. Use it before you hit enter and your AI output gets dramatically better.
The 5 questions
1. What's the output? (Be specific.)
Bad: "Write me a proposal." Better: "Write me a 4-paragraph email proposal." Best: "Write me a 4-paragraph email proposal in the style of the example below, addressed to the buyer named Sarah, for the project described below, ending with a single clear next-step CTA."
The more specific the output, the better the result. AI defaults to a generic average if you don't give it parameters. Always tell it exactly what shape the output should be.
2. Who's it for?
The AI doesn't know your audience unless you tell it. Same content for "a CFO at a 200-person SaaS company" reads completely differently than "a busy office manager at a 12-person dental practice."
Specify the audience: their role, what they care about, what they don't care about, what they already know. Two sentences is enough. "This is for a 12-person dental practice owner who knows their patient management system but doesn't know AI vocabulary. They care about ROI and team morale. Skip the technical detail."
3. What's your voice?
If you don't tell the AI what voice to use, it will write in default-corporate-blog voice. That's the voice everyone hates. The voice everyone hates is also AI's natural setting.
Three ways to specify voice: - "Write in the style of [link to past example]" - "Write like [name of writer whose voice you want]" - "Write in this voice: [paste 200 words you wrote]"
The last one is the strongest. AI matches voice surprisingly well from a sample. Drop in two paragraphs of your past writing and it'll usually nail the register.
4. What does GOOD look like?
This is the question most people skip. AI is great at imitation and bad at imagination. If you can show it an example of "this is what I want it to look like when it's right," you'll get much closer on the first try.
Even one paste of "here's a similar piece I think is good, write something in that shape but for [new context]" works better than describing it.
If you don't have an example handy: write 3 bullet points of what makes the output good. "Good output: starts with the buyer's specific problem, doesn't lead with our credentials, includes one specific number, ends with a question that's easy to say yes to."
5. What does BAD look like?
Tell it what to avoid. AI listens.
"Don't use the word 'leverage.' Don't open with 'In today's...'. Don't include disclaimers. Don't use bullet points unless I asked for them. Don't write more than 200 words."
Negative constraints are surprisingly powerful. AI defaults to a lot of bad habits (corporate jargon, hedging, over-formatting). Tell it which ones to drop and the output gets cleaner immediately.
The full template, assembled
Put the 5 answers together in any order. A typical good prompt:
> Write a 200-word email proposal. It's going to Sarah, the office manager at a 12-person dental practice. She's tech-comfortable but not a technical buyer. The voice should be operator-direct and slightly casual — match the example below. Good output starts with her specific pain (front-desk overload at 4:30pm) and ends with a single clear yes/no question. Don't use "leverage," don't open with "In today's," don't pad with disclaimers. > > [Paste example email + paste relevant context]
That's it. Five questions answered in 100 words. The output you'll get is dramatically better than the default.
What NOT to bother with
Three pieces of common prompt advice you can skip:
1. "You are an expert in..." preambles. The AI doesn't believe these and they don't change the output much. If you want expertise, give it real source material to reference.
2. ALL CAPS WARNINGS. Doesn't help. The AI still does the thing you said not to do at the same rate.
3. Multi-page prompt frameworks copied from Twitter. Most of them are over-engineered. The 5 questions cover what you need.
When the output is still bad
Two debugging moves:
Move 1: Ask why. Type back: "That output isn't quite right. The voice is too corporate and it didn't address Sarah's specific pain. What about the prompt would have produced a better result?" The AI will tell you what context it was missing.
Move 2: Show it the better version. Edit the AI's output yourself, paste your edited version back, say "this is the version I want — what should I change in my prompt next time to get closer to this on the first try?" The AI will give you specific prompt edits.
Most people stop at "the output was bad" and blame the AI. Half the time the prompt was the problem and a 30-second debug saves you 10 minutes of editing every future prompt.
What this means for you
Prompts aren't magic spells. They're instructions. The clearer the instructions, the better the output. The 5 questions force you to be clear in 30 seconds, instead of accepting the default.
Test it on the next AI task you do. Compare 30 seconds of prompt prep vs. zero seconds. The output will be visibly better.
Get started
Want to know what we'd build for your team?
30-minute audit walks through your workflows and outputs a list of 2-3 candidates ranked by ROI, with prices.