Getting started

How to Write AI Prompts That Actually Work

By AskGeeks.ai · · 7 min read

Hands writing on a sheet of paper at a warm wooden desk beside an open laptop.

A good business prompt contains four things: the role or context, the task, the audience and constraints, and an example of what good looks like. Vague prompts produce vague output, and almost every complaint about AI writing "sounding generic" traces back to a prompt that never said who it was for.

The structure

  • Context — who you are and the situation. "I run a two-van plumbing business in Vancouver."
  • Task — what you want, specifically. "Write a follow-up email to a customer who received a quote nine days ago and has not replied."
  • Audience and constraints — who reads it, how long, what tone, what to avoid. "Friendly but not pushy, under 120 words, no discount offer."
  • Example — paste something you have written that you liked. This matters more than everything else combined.

Before and after

Weak prompt

"Write a follow-up email to a customer."

Produces something polite, generic and faintly corporate that you will rewrite entirely. The model had nothing to work with.

Strong prompt

"I run a two-van plumbing business in Vancouver. Write a follow-up to a customer who got a quote nine days ago and hasn't replied. Friendly, not pushy, under 120 words, no discount. Here's an email I sent last month that worked well: [paste]."

Produces something you can send after a small edit, in your voice rather than a template's.

Five habits that do the heavy lifting

  • Show, don't describe, your tone. One pasted example beats ten adjectives.
  • Ask it to interview you. "Before writing, ask me any questions you need." Excellent for anything complex.
  • Iterate in small steps. "Shorter. Warmer. Cut the last line." Do not restart from scratch.
  • Ask for options. "Give me three versions with different angles" — picking beats editing.
  • Make it critique itself. "What is weak about this draft?" often surfaces the exact thing that was bothering you.
Treat it like briefing a capable new hire who has never met your customers. You would not say "write a follow-up email" and walk away.

Prompts worth saving

The real productivity gain is not writing better one-off prompts — it is keeping the handful that work. Most small businesses need perhaps six: a quote follow-up, an enquiry reply, a job ad, a service description, a review response, and a summarise-this-thread prompt. Save them in a shared document with the placeholders marked, and the whole team gets the benefit rather than one person who happens to be good at it.

Where prompting stops helping

Prompting is how you get better output from a chat window. It does not make the tool aware of your bookings, and it does not make anything happen automatically. When the job is "this should run without anyone opening a browser tab", you have moved from prompting to building — see AI agent vs AI chatbot.

And whatever you prompt, keep confidential material out of public tools — what you should never tell ChatGPT covers the specifics. If you would like help identifying which workflow is worth automating properly, the free AI Opportunity Scan does exactly that.

Questions

Common questions

What is an example of a good prompt?

A good prompt supplies context, task, audience and an example. For instance: "I run a two-van plumbing business in Vancouver. Write a follow-up email to a customer who received a quote nine days ago and hasn't replied. Friendly but not pushy, under 120 words, no discount offer. Here is an email I sent last month that worked well: [paste]." Compare that with "write a follow-up email to a customer", which gives the model nothing to work with and produces generic output you will rewrite anyway.

How do you structure a prompt?

Use four parts. First, context — who you are and the situation. Second, the task stated specifically. Third, audience and constraints such as tone, length and anything to avoid. Fourth, an example of output you liked, which improves results more than any other single change. For complex requests, add a fifth instruction telling the model to ask you clarifying questions before it answers, so it gathers what it needs instead of guessing.

Why does AI writing sound generic?

Almost always because the prompt did not say who the writing is for or show what good looks like. With no audience, no constraints and no example, the model produces the safest average of everything it has seen, which reads as bland corporate text. Fix it by pasting a sample of your own writing, naming the reader, setting a length limit, and stating what to avoid. Then iterate in small steps rather than rewriting the request from scratch.

Keep reading

More from the blog

A person watching a task complete by itself on a laptop at a sunlit desk.
AI basics · 6 min read

AI Agent vs AI Chatbot: What's the Difference?

A chatbot has a conversation. An AI agent takes actions across your systems and completes a task end to end. That difference decides whether you are buying a tool that deflects questions or one that removes work — and most businesses buy the wrong one.

Read the post →

A small brass padlock resting on a closed folder beside a laptop in warm light.
Safety & governance · 7 min read

Is Your Business Data Safe With AI?

The risk is not AI in the abstract — it's an employee pasting confidential data into a public tool whose terms let it be stored. Safety is a set of choices, not a property of the technology. Here's how to make the right ones.

Read the post →

Get started

Find the first AI system worth building.

Show us where your business is losing time, revenue, capacity or visibility. We will help determine whether AI can solve it — and what the first practical step should be.

Book a Free Opportunity Scan See the Starter Pack

hello@askgeeks.ai · Vancouver, BC · remote across Canada & the US