How to Write Effective AI Prompts: A Simple Formula and 15 Examples for Everyday Work

Quick story first. A few months back I asked ChatGPT to write a product description for a stainless tumbler I was selling. What I got: “perfect for any lifestyle.” A tumbler. For hiking. I don’t hike, and neither do my customers. That sentence could’ve been selling a watch, coffee beans, or life insurance.

So the problem wasn’t the AI. It was me typing “write me a product description” and expecting magic. That’s wishing, not instructing.

Why the answers come out bland

Understand one mechanism and everything clicks. These models are next-word guessing machines. Give them a vague question and you get the statistical center of everything ever written on the topic. Polished. Competent. Lifeless.

And weak prompts almost always have three holes: no writer assigned, no context about your actual situation, no shape for the answer. Email? Table? Three quick options? Without that it guesses, and it always guesses the safest, most boring version.

The formula I landed on: Role + Task + Context + Format

Four pieces. I cobbled it together from different places and simplified it myself, and honestly it works almost every time.

Role is who’s writing: “Act as an email copywriter.” Task is the job: “write a re-engagement email.” Context is what it can’t possibly know: “customers who haven’t ordered from my shoe store in 90 days.” Format is the shape: “under 120 words, friendly, one button at the end.”

Watch the difference:

“Write a sales email.”

Versus:

“Act as an email copywriter. Write a 120-word re-engagement email for customers who haven’t ordered from our online shoe store in 90 days. Friendly tone, mention the new sandal collection, one clear call-to-action button.”

Ten extra seconds to write. Twenty minutes of editing saved. I timed it once with a stopwatch out of spite, the gap was ridiculous.

15 prompts stolen from my own saved file

These actually live on my laptop, part of a forty-something prompt collection. The ones below get used most, just swap the brackets.

Emails

1. “Act as a customer service lead. Write a polite reply to a customer asking for a refund on a [product] that arrived damaged. Offer a replacement first, refund second. Under 100 words, warm tone.”
2. “Write a follow-up email to a potential client who got my quote for [service] five days ago and went quiet. Non-pushy, one question to reopen the conversation.”
3. “Draft a payment reminder for invoice [#], 7 days overdue. Firm but friendly, mention the amount [Rp amount] and a payment link.”

Reports and documents

4. “Summarize this sales report into 5 bullet points for my business partner, focusing on what changed versus last month: [paste report].”
5. “Turn these meeting notes into decisions, action items, owners, deadlines: [paste notes].”
6. “One-paragraph executive summary of this proposal, for a non-technical reader: [paste proposal].”

Summaries and reading

7. “Summarize this article in 3 sentences, then list 2 points relevant to a small online business in Indonesia: [paste article].”
8. “Explain this contract clause in plain language for a freelancer, and flag anything risky: [paste clause].”

Analysis and decisions

9. “Here are my sales numbers for the last 6 months: [paste data]. Find the 3 biggest trends, suggest one action each.”
10. “Compare option A and option B for [decision]. Table with cost, time required, biggest risk. Then recommend one.”
11. “Review this product description. Give me 3 reasons a customer might hesitate, plus a fix for each: [paste description].”

Content and marketing

12. “Write 5 Instagram caption ideas for a [product] launch. Under 20 words each, conversational, end with a question.”
13. “Generate 10 blog title ideas about [topic] for beginners. Long-tail, under 60 characters.”
14. “Rewrite this paragraph for a landing page. Under 50 words, lead with the customer benefit: [paste text].”
15. “Make a 7-day Instagram content plan for a small [type of business]. Mix education, product, behind-the-scenes.”

Mistakes I still keep making

Cramming five requests into one prompt. Every time I do it, I get five half-baked outputs instead of one good one, and every time I’m annoyed at myself. Split them up. Seriously.

Then forgetting the audience. “Explain inflation” versus “explain inflation to a warung owner who’s never thought about it” get wildly different answers, yet I still skip it when I’m rushing.

And the big one: accepting the first draft. The first output is a starting point, period. The good stuff shows up when you nudge it, “shorter,” “less stiff,” “give me an example.” One push at a time, not a full rewrite demand.

Oh, and the funny one: asking it to summarize a file I never pasted. It can’t read your laptop. What it does instead is invent things, and unfortunately it invents them with total confidence.

Check before you trust

It writes with complete confidence while fabricating, prices, dates, regulations, statistics. Verify everything against the official source before it goes live. Read the output out loud; anything that sounds like a press release, delete. And if it mentions a feature or quote you never provided, treat it as fake until proven real. Been burned, it once cited a “2023 study” that never existed anywhere.

Your first move

Sticky note. The four formula pieces. Monitor edge. Leave it there until you stop needing to glance at it, took me about three weeks.

Then grab one task you do weekly. A report. A client email. A caption. Run it through the formula once, compare it against the lazy one-liner result.

Everything that works goes into one document. Mine’s at forty-something prompts now, collected over two casual months. Somewhere around week three the AI stopped feeling like a slot machine and started feeling like a fairly quick assistant who just needs painfully clear instructions. That’s the whole secret, really. Not a fancier tool. Clearer instructions.