Module 3 of StartCloud's AI at Work: Foundations learning pathway, in five short units with a knowledge check: why vague prompts produce generic answers, the four parts of a clear request (goal, context, format, tone), how to refine a first draft through follow-ups, and five ready-made prompt patterns for everyday work in any AI tool.
Getting Better Answers From AI
Vague in, vague out
Type "write me a report" into any AI tool, ChatGPT, Copilot, Claude, Gemini, take your pick, and you will get the same thing back: three tidy pages that say nothing. An opening about challenges and opportunities. A middle that hedges. A closing that recommends further discussion. Beige, wall to wall.
Here is why. The tool cannot read your mind. It knows nothing about your customers, your prices, or the reason you need this report by Friday. Given nothing to work with, it does the only thing it can: it averages across every report ever written and serves up the safest middle of the lot. The blandness is not a fault in the tool. It is a faithful reflection of the question.
Which is genuinely good news. The person in your office who gets great answers out of AI is not smarter than you and does not have a better version of the tool. They just put more into the question. That is the entire skill, and this module hands it to you in about eight minutes.
"Write me a report on the quarter."
What comes back: confident waffle. Vague headings, filler about market conditions, and not one number or name from your actual business, because you gave it none. You then spend twenty minutes rewriting it, and quietly decide AI is overrated.
"Write a one-page report on our June quarter for the two owners. Sales were up 8 percent, two staff left, and the new booking system went live in May. Focus on what changed and what needs a decision. Plain English, short paragraphs."
What comes back: a draft built from your facts, shaped for your readers, that you edit rather than rewrite. Thirty seconds of typing bought you most of an hour.
A search engine rewards short and punchy, so twenty years of Google trained us all to type three words and hope. AI tools reward the opposite. They will happily use every scrap of detail you give them, and they cannot invent the scraps you leave out. If the answer is generic, the fix is almost never a better tool. It is a better question.
You do not need a prompting course, a cheat sheet of magic words, or any of the folklore floating around LinkedIn. A useful request answers four questions: what you want, what the tool needs to know, what shape the answer should take, and how it should sound. Goal, context, format, tone.
You will not need all four every time, though skipping the goal is how beige gets made. Each extra detail narrows the tool's guessing, and the answer improves in direct proportion. It works the same in every AI tool, because underneath, they all respond to the same thing: being told.
Name the thing. An email, a summary, a list of options, a set of questions, a rewrite. Start with a verb and be specific about the job, not just the topic.
Who the answer is for, what is going on, and the key facts. This is the part people skip most, and it is the one that turns filler into something that sounds like your business.
Length, structure, bullet points or paragraphs, a table, a subject line included or not. If you can picture the finished thing, describe the picture.
Friendly, formal, direct, apologetic, upbeat. Two or three words is plenty. Without them the tool defaults to a polite corporate hum that sounds like nobody in particular.
All four, in one everyday request
Say a benchtop supplier has pushed a delivery back and a customer needs to hear it from you. Here is the full request assembled, with each part labelled.
Write a short email to a customer letting her know her kitchen benchtop install has moved back a week. The stone supplier pushed delivery to the 18th, so the soonest we can install is the 20th. She has already been patient through one earlier delay, and we can lock the new date in today. Three short paragraphs, new date stated clearly in the first one, and end with a direct line to call if the 20th does not suit. Warm and straight. Apologise once, properly, without grovelling.
Notice there is nothing clever in there. It reads like a normal instruction to a capable colleague, which is exactly the register to aim for.
One caution before you get generous with context. Real client names, personal details, and anything confidential need care before they go into any chatbot, and Module 4 of this pathway covers exactly what should stay out and why.
Most people use AI like a vending machine. Coin in, snack out, walk away, and if the snack is disappointing, oh well. That habit throws away the tool's best feature: it remembers the conversation you are in, so the first answer is not the product. It is the opening line.
Nobody expects a colleague's first draft to be the final one either. You would hand it back with a note: tighten this, soften that, lose the second paragraph. Do the same here. Follow-ups cost you five words and five seconds, and they are where the quality actually happens.
- "Shorter."
- "Warmer, this reads a bit stiff."
- "More direct. Get to the point in the first sentence."
- "Give me three versions that take different approaches."
- "Half the length, keep every fact."
- "Now do it again for someone who has no background on this."
No need to repeat the whole request. The tool still has it. Just say what to change, the way you would across a desk.
Three moves that feel like cheating
Before you fix a draft yourself, try: "Critique that draft. What is weak, what is missing, and what would a sceptical reader push back on?" Then ask it to redo the draft with its own criticisms fixed. The second version is usually a clear step up, and you did none of the lifting.
When you are not sure what detail matters, hand over the steering wheel: "Before you answer, ask me five questions that would help you do this well." Its questions will surface context you would never have thought to volunteer, and the eventual answer is built on all of it.
Paste in a paragraph or two you have written yourself, an old email is perfect, and say: "Match this style." From then on, drafts come back sounding like you rather than like a press release, which saves the most tedious editing pass of all.
The pattern behind all of this: stay in the chat. Two or three rounds of back and forth beats ten attempts at writing the perfect prompt on the first go, every single time.
Everything so far becomes automatic with practice, but practice needs starting points. Here are five patterns that earn their keep in almost any job, in any AI tool. Swap the square brackets for your own details, and remember the caution from earlier about what real client information goes in.
"Here is an email thread. Summarise what has been decided so far, then list who owes what to whom, with any dates or amounts mentioned."
"Turn these rough dot points into a short written update for [who will read it]. Keep every fact exactly as given, add nothing new, and flag anything that looks incomplete."
"Rewrite this explanation for a customer who is not technical. No jargon, keep it friendly, and do not change what actually happened."
"Here is the agenda for a meeting I am attending tomorrow. Suggest the questions I should ask for each item, and anything I should read or prepare beforehand."
"Explain this document back to me in plain English, section by section. Then list anything a careful reader would want clarified before agreeing to it."
Everything in this module applies there too, and Copilot adds a twist: it can ground its answers in your own files, emails, and meetings. Our Copilot Essentials pathway goes deeper on prompts that draw on your business data directly.
- Google Workspace: Writing effective AI prompts for business
- Australian Cyber Security Centre: Engaging with artificial intelligence
Guidance was current at the time of writing (July 2026). The tools move quickly, but the habit of asking well has stayed remarkably stable while everything around it changed.
Knowledge check
5 quick questions. Get 4 right and the module is yours.
1. Why does a prompt like "write me a report" produce generic filler?
2. What are the four parts of a useful AI request covered in this module?
3. You type "write an email to a client about the delay" and get something bland. Which single change would most improve the answer?
4. The first draft comes back stiff and formal when you wanted friendly. What is the best next move?
5. You need to brief the tool on a tricky situation but are not sure which details matter. Which move from this module fits best?