Before and after: 5 real prompts, rewritten

Principles click faster with real cases. Here are five requests people make to AI every day, before and after filling in the missing layers, with a note on what changed and why.

1. Replying to a refund request

Before

Write a reply to a customer who wants a refund.

After

You are a customer support lead for a small online shop.

I sell handmade ceramic mugs. A customer says their mug arrived
chipped and wants a refund. Our policy: full refund or free
replacement for damage in shipping, no need to return the item.
This customer has ordered from us 4 times.

Write a reply that apologizes, offers both options and asks which
they prefer. Warm and brief, under 90 words. No corporate phrases
like "we apologize for any inconvenience."

What changed: the policy and the customer history set the tone; the banned phrase shows exactly what “not corporate” means.

2. A website for a small business

Before

Build me a website for my dog-grooming business.

After (for Claude Code)

<context>
I own a dog-grooming salon. I can't code. I want a one-page site
so people can find me on Google and book by text message.
The folder is empty; this is a new project.
</context>

<task>
Build the site with: name and tagline, services with prices,
a 6-photo gallery, hours, a map, and a "Text to book" button
that stays visible on phones.
</task>

<constraints>
Plain HTML and CSS, no frameworks. It must be free to host.
</constraints>

<how_to_work>
Propose a plan first and wait for my approval. Then tell me how
to open the site on my computer to see it.
</how_to_work>

What changed: skill level, exact content and a way of working. More in Claude Code prompts.

3. Analyzing a spreadsheet

Before

Analyze this sales data.

After

You are a data analyst for independent retailers.

Attached is 2025 sales for my bike shop, by month and category.
I need to decide which categories to cut back so I can stock
more e-bikes.

1. Rank categories by revenue and by margin.
2. Flag any that dropped more than 20% from 2024.
3. Recommend two categories to shrink, with the reason.

Use only the data provided. If something's missing, say so.
Output: one table, then three one-line findings.

What changed: the decision behind the analysis, plus concrete steps.

4. A support bot

Before

Make a bot to answer my customers.

After: a system prompt covering identity, business facts, what the bot does, what it never does, and when to hand off to a human. The full walkthrough is in how to write a system prompt for a customer service bot.

5. Studying a topic

Before

Explain compound interest.

After

You are a personal finance teacher who uses everyday examples.

I'm 24 and just opened my first savings account. I'm not good
with math.

Explain compound interest in under 150 words using a savings
example with real numbers. Then ask me 2 questions to check I
understood, and wait for my answers before correcting me.

What changed: the learner’s level and an interactive step that turns an explanation into a lesson.

The pattern

Every improved prompt answers the same questions: who’s helping, in what situation, what to deliver, in what shape, within which limits. That’s the anatomy of a great prompt. To build yours without starting from a blank page, try the coach.