A good business prompt has five parts: the role the AI should play, the exact task, the context it can’t guess, an example of good output, and the format you want back. You won’t always need all five, but the more you include, the better the first result.
- Most bad AI results come from vague instructions, not bad tools.
- Use the recipe: role, task, context, example, format.
- Paste in examples of work you liked.
- Save prompts that work as templates for the team.
Why AI gives generic answers
If you ask “write a reply to this customer”, the AI has to guess your tone, your policy, how long the reply should be and who is speaking. It fills those gaps with the most generic option. The fix isn’t a better tool. It’s twenty extra seconds of instruction.
The 5-part recipe
Weak vs strong: the same request, twice
- “Write a reply to this customer.”
- AI guesses the tone, policy and length
- Generic result you rewrite from scratch
- “You’re our front desk assistant. Write a reply to this customer asking about a late delivery. Our policy is a free reschedule. Under 100 words, warm and plain, signed ‘the team at Bright Plumbing’.”
- Specific, on-policy, on-brand
- Send after one quick check
Five prompts you can copy today
You are our office manager. Summarize this email thread in 5 bullet points: what the customer wants, what we promised, open questions, deadlines, and the next step. Plain English, no jargon. [paste thread]
You are our marketing assistant. Using the voice guide below, turn this one idea into: 1 LinkedIn post (under 150 words), 1 Instagram caption, and 1 short email newsletter section. No exclamation marks, no claims we can’t prove. Voice guide: [paste] Idea: [paste]
You are our bookkeeper. From this invoice, extract: supplier, invoice number, date, due date, each line item with quantity and price, tax and total. If anything is unclear or doesn’t add up, say so instead of guessing. [paste invoice text]
You are our front desk assistant. Draft a reply to this review. Thank them by name, mention one specific thing they said, and invite them back. Under 60 words. Don’t mention discounts or staff names. [paste review]
You are a careful editor. Check this draft for wrong facts, numbers that don’t add up, wrong names, and anything that sounds like a promise we can’t keep. List each problem with a suggested fix. [paste draft]
Three ways to teach AI your way of doing things
- Show examples. Paste two or three past emails or documents you were happy with and say “write it like these”.
- Share your words. List the phrases your business always uses, and the ones it never does.
- Save what works. When a prompt gives a great result, save it as a template so the whole team can reuse it.
Notice the last line of the invoice prompt: “say so instead of guessing.” Asking AI to flag uncertainty is one of the simplest ways to reduce made-up answers.
Before you paste anything sensitive into a prompt, read what you should never share with an AI tool. And to make prompting a team habit, see how to get your team using AI.
How to fix a disappointing result
Even good prompts sometimes miss. Instead of starting over, tell the AI exactly what to change, the way you would with a colleague. Specific feedback gets specific improvements.
Common mistakes to avoid
One clear task per prompt beats a paragraph of mixed requests.
Say who will read it: a customer, a supplier, your accountant.
Paste the email, invoice or notes. AI can’t check what it can’t see.
Facts, numbers, names, tone. Every time, before it leaves your hands.
When we set AI up for a client, the instructions behind it look a lot like this recipe, just longer: who the AI is acting as, what it must do, the business rules it can’t guess, examples of good work, and the exact format we need. Clear instructions are the cheapest quality control there is.
Where to go from here
The fastest way to learn this is to try it on one real task this week. If you want a guided start, take the free course. If you’d rather talk it through, book a call.
Questions, answered
What makes a good AI prompt for business?
Clarity about who the AI is acting as, the exact task, the context it can’t guess, an example of good output, and the format you want. That’s the 5-part recipe.
Do I need to be technical to write prompts?
No. If you can brief a new colleague by email, you can write a good prompt.
How do I stop AI from making things up?
Give it the source material, ask it to say when it’s unsure instead of guessing, and check facts, numbers and names before anything goes out.
Should my team share prompts?
Yes. Save prompts that work as templates so everyone gets consistent, on-brand results.