A prompt isn't a Google search. It's how you describe the job to a very fast colleague — one who fills in the blanks when you leave gaps.
Good news: you don't need technical vocabulary to get better answers. Six simple approaches cover most everyday work in a small or mid-size business. Each one takes about two minutes to try in whatever tool you already use.
At a glance
- Direct ask — when the task is simple and unambiguous
- Show one or two examples — when format or judgment matters
- Step by step — when several factors are mixed together
- Break it into pieces — when the input is long or spans multiple themes
- Ask for a critical reread — before sending anything to a client
- Set the role up front — so tone and depth stay right
Two layers: frame and request
Before the six techniques, one useful distinction:
| Layer | What it is | Example |
|---|---|---|
| Frame (role + rules) | How the AI should behave in general | "You write for a 30-person professional services firm in Quebec. Use formal address with clients. Never promise a deadline without a specific date." |
| Request (today's task) | What you want right now | "Classify this email: urgent / normal / can wait." |
The frame can stay the same in a team template. The request changes every time. The six techniques below mostly apply to the request — plus business context when it matters.
1 — Direct ask (simple task)
When to use it: obvious classification, short definition, low-stakes rewrite.
Example — triage inbox
Classify this message as Urgent / Normal / Can wait.
One-sentence reason.
Message:
"Hi — our May 12 invoice doesn't match the purchase order.
Can someone call back before end of week?"
Expected result: Urgent — billing mismatch + implicit deadline.
Limit: if the message is ambiguous or sensitive, add context (key account, dispute history) or use a richer technique.
2 — Show one or two examples
When to use it: the model needs to match a format, tone, or judgment that words alone don't nail.
Example — label customer comments
You label comments for our support team.
Labels: Positive | Neutral | Negative | Mixed
Examples:
1. "Fast delivery, careful packaging." → Positive
2. "Product is fine, nothing special." → Neutral
3. "Good product, but support took five days to reply." → Mixed
Now label this one:
"Very happy with the report, but layout took two rounds of revisions."
Why it works: "Mixed" isn't always obvious without examples. Two or three cases are enough — not a full page.
3 — Reason step by step
When to use it: mixed sentiment, multi-criteria decisions, analysis where a one-word answer would mislead.
Example — nuanced customer comment
Analyze this comment step by step, then give a final verdict.
Comment:
"We got the report on time. The analysis is solid.
However, two charts still used last year's numbers."
Steps:
1. What is clearly positive?
2. What is a problem?
3. Is the problem major or minor for the client relationship?
4. Overall verdict: Positive | Mixed | Negative — and why in 2 sentences.
Expected result: Mixed — satisfied with substance, specific quality issue on data.
Tip: asking for steps before the verdict cuts down on answers that are too rosy or too harsh.
4 — Break it into pieces
When to use it: long text, several themes, or output that different people need in different shapes.
Example — 45-minute meeting notes
Split the notes below into three separate blocks.
Do not blend blocks together.
Block A — Decisions made (bullets)
Block B — Actions: who | what | due (table)
Block C — Open items + next decision date
Rule: do not invent names or dates missing from the notes.
[… paste notes …]
How this differs from step by step: here you structure the output up front. The AI doesn't "summarize in a blob" — it fills slots your team already uses.
5 — Ask for a critical reread
When to use it: drafts for clients, executives, or any high-stakes message.
Example — client email after a delay
Step A — produce the draft (with context, like our before/after context examples):
[… prompt with facts, tone, constraints …]
Write subject line and body.
Step B — critique (new message or follow-up):
Reread the draft above as a cautious operations director.
List:
1. Implicit or explicit promises we cannot keep
2. Tone too cold or too casual for this client
3. Facts not supported by my original context
Propose a revised version fixing those points.
Cost: one extra minute. Benefit: fewer "oops" after Send — especially with human review before the final click.
6 — Set the role up front
When to use it: almost always for recurring work — even when today's task looks simple.
Example — reusable template
You are a writing assistant for a 35-person firm (consultants + coordinators), Quebec.
Standing rules:
- Canadian French or English as I specify; formal with clients unless noted
- No empty hype ("revolutionary solution")
- If information is missing, write [TO CONFIRM] instead of inventing
- Default length: short unless I say otherwise
Today's task:
Write a polite payment reminder — invoice 30 days overdue, known client, no legal threats.
In practice: save the "role + rules" block in a shared doc. Paste it at the start of each session — or use a tool that keeps it in memory.
Which technique when?
| Situation | Start with |
|---|---|
| "Is this urgent?" on one message | Direct ask |
| Same format every week | Examples + framed role |
| Mixed customer feedback | Step by step |
| Long meeting notes | Break into pieces |
| Client email or proposal | Draft + critical reread |
| New teammate using AI | Framed role + team templates (leader tips) |
Common mistake: stacking all six techniques on a three-line task. Email triage doesn't need a seven-step journey.
Five-minute drill
Take a real internal email (sanitized). Try in order:
- Direct ask — "Summarize in 3 bullets."
- If the summary is vague — add role + context in 3 lines
- If the topic spans two departments — split into Decisions / Actions / Open questions
Compare the three outputs. The gap teaches more than theory.
Where you are
You now have six practical levers — without academic jargon. To standardize this as a team, see Prompting tips for leaders and teams. For the business context that prompts often miss, see Why context in your prompt is crucial.
If AI drafts sound polished but miss how work actually runs, Let's talk. We'll build two or three templates on your real workflows — not generic examples.