ATER Studio ↗ไทย
AI & Work Culture

The prompt that suits me best.

The story of Mali, a white cat named Cotton, and two AIs with very different jobs

13 November 2026 · 8 min read

A Folder Full of Other People's Work

On Thursday night, Mali opened the folder called "Secret Prompts, Don't Miss Out" yet again.
Inside were thirty-odd files: "Prompt for Product Captions That Pop," "Prompt for a 30-Day Content Plan," and "The Prompt That Makes AI Think Like a Marketing Consultant."

She had saved every one of them from a free giveaway post.
Mali runs a small ceramics shop. She makes coffee cups by hand, one at a time, at home. Most of her customers love handmade things and are happy to wait. She had been pasting the prompts in one by one. Every result read fine, line by line, but not one line sounded like her.

"This cup will elevate your every morning ✨" the AI wrote.

Cotton, Mali's white cat, walked across the keyboard and sat down on her hand, as if to say it was getting late.

Mali looked at the screen and asked herself, quietly : who were these prompts written for? Not for her, that much was clear. Whoever wrote them didn't know she threw every cup herself, didn't know who her customers were, and didn't know she couldn't stand the word "elevate."


Chapter 1: A Shirt Cut for Someone Else

The next day Mali told her sister, who works in HR. Her sister laughed and offered a comparison: a ready-made prompt is like a free-size shirt. It fits everyone, and fits no one well.

And the poor fit isn't because the people sharing them aren't good. Some are very good.
But their best prompts were cut from their own context: their business, their customers, their voice, their constraints. Drop one into our life and the gaps get filled with the average of everyone in the world, which is the kind of blandness anyone can tell was written by AI.

So Mali asked herself one question. What if I stopped looking for the best prompt, and looked instead for a way to build the prompt that fits me best? Where would I even begin?

The answer turned out to be simple. If the context lives in my head, then the one who should be asked is me, not the AI.

Questions to chew on
• Think of the last prompt you saved. Did you ever know what kind of work its author used it for?
• If you had to explain yourself to a stranger in five sentences, what would you say first?


Chapter 2: The First AI, an Interviewer Who Isn't an Expert

Mali opened a fresh chat window and decided that this window would "produce nothing at all." It had exactly one job: to ask questions. She typed:

You are an interviewer. Your job is to help me find out what I need from another AI.

My big goal is: [briefly describe the result you want, e.g. I want captions for my ceramic cups that sound like I wrote them myself]

Rules:
1. Ask only ONE question at a time, and wait for my answer before asking the next.
2. Start with broad questions about the result I want, then move into detail.
3. Cover: who will read or use the result / the tone I like and dislike / examples of work I love or hate / constraints (time, tools, budget, language) / what I've tried that didn't work / what "a job well done" looks like.
4. If my answer is vague, ask a follow-up instead of guessing.
5. Do not suggest solutions or write anything for me yet.
6. When you think you have enough, tell me you're ready to summarize, and ask whether there's anything I'd like to add.

Then the interview began. The first question was disarmingly simple: "Who was your most recent customer, and what did they say about why they bought?"

Mali typed her answer and then paused, because she had just realized that none of her last three customers had mentioned "quality."
Every one of them had talked about "the fingerprint at the bottom of the cup."
That detail had never appeared in a single prompt in her folder.

Questions to chew on
• A good question doesn't hand us a new answer. It lets us see what we already knew but had never said out loud. Have we had a moment like that?
• Why is having the AI ask one question at a time different from sitting down and writing out every detail ourselves in one go?

Chapter 3: When the Answers Become a Map
After a dozen or so answers, the interviewer AI said it was ready to summarize.
Mali chose not to close the chat. Instead she typed one more message, which is the heart of the whole method.

From the whole interview, please do two things for me.

Part 1: Context Brief
Write a short document that another AI, one that doesn't know me at all, can read and understand. Include: who I am and what I do / who my readers or customers are / the tone that fits and the tone to avoid (with examples from what I told you) / constraints / what counts as success.

Part 2: A step-by-step plan
Break the work into small steps, in order. For each step give me:
- The name of the step and its goal
- A prompt I can copy and paste right away (complete on its own, and allowed to refer to the Context Brief)
- What I should check before moving to the next step

Write everything so I can copy it into a different AI window.

What came back wasn't a single prompt. It was a structured document.
At the top was a context brief telling Mali's story. Below it were five steps: gather the words customers use to describe the cups, write the story outline of a caption, draft three versions, adjust to her tone, and check for words she dislikes.
Each step had its own prompt and its own checklist at the end.

Mali read it and realized this was exactly what had been missing from her folder all along.
Every prompt in it told the AI how to do something. Not one said for whom.
Cotton jumped onto the desk and batted a pen onto the floor, which Mali chose to interpret as a score.

Questions to chew on
• If we let someone who doesn't know us read only our context brief, how much of us would they understand?
• Take the work we want AI to do and break it into small steps. Is there a step we ourselves had never thought of as a step?


Chapter 4: The Second AI, a Doer Who Doesn't Have to Guess

Mali opened a new window, not a continuation of the old one.
She wanted this AI to start from zero, with only what she had chosen to give it, and without a long interview lingering in its memory.
She pasted in the context brief first, then typed something short:

This is my context. Read it, then tell me briefly what you understand about me and this work. After that, wait for me. I'll send the work one step at a time.

Having the AI say what it understands before it starts is a small trick that helps a lot, because if it has misunderstood, you find out before spending any effort on step one.
The AI replied: "You shape each cup by hand. Your customers value the traces of a human hand, and you dislike inflated advertising language." Mali smiled. It was right on every point.

Then she worked through the steps one at a time: paste the prompt for step one, read the result, check it against the list at the end of the step, and only then move on.
At step three she found a draft caption that was still a little stiff, and told it plainly: "This still sounds like a brochure.
Try writing it as if you're telling a friend." The AI adjusted at once, because it now knew who Mali's friend was.

Questions to chew on
• How is separating the one who asks from the one who does different from having a single AI do everything? In real life, do we split those two roles in other kinds of work?
• Which steps in our work should we decide ourselves every time, no matter how good the AI is?


Chapter 5: When the Result Isn't Right Yet, Go Back and Ask Again

A week later Mali ran into a new problem. She had opened an indigo-blue cup collection, something she had never made before, and the captions that came back had that same old flavor again.

She didn't go back to the folder of secret prompts. She went back to the interviewer AI and said: "There's something new. Ask me only about the parts that changed, then update the context brief."

Three questions later, she saw that the indigo customers were a different group from her usual ones. They liked stillness, not warmth. She got a second version of the context brief, and prompts adjusted to match.

This is where Mali understood that what she had built wasn't a prompt. It was a living document. Her business changes, her customers change, she herself changes, and the document simply has to be asked again to keep up with her.

Questions to chew on
• If the context brief we wrote a year ago is still in use, is it still about us?
• How would we know when it's time to have the AI interview us again?


Conclusion: What Works for Everyone Isn't a Prompt. It's the Questions.

If we boiled Mali's story down to a few lines, it would be three ideas.

1. A good prompt comes from context. Context lives in our heads, not in someone else's giveaway post.
2. Separate asking from doing. One AI draws out what we know. The other works from what has been summarized.
3. Break the work into steps and check each one. Results get better when we can see them along the way, not just at the finish line.

The prompts other people give away aren't wrong. They can be inspiration and good examples of structure. But if we start from them, we're starting from someone else's answers and only later looking for our own questions.
Mali's way runs the other direction: questions first.
That night Mali deleted the folder "Secret Prompts, Don't Miss Out." No, she didn't delete it. She just renamed it "Ideas to Browse for Fun" and opened a new file called "My Context, Latest Version."
Cotton curled up beside the screen, eyelids heavy, its white fur turning pale blue in the glow. Mali typed the first line of the next question she would ask herself, and smiled.


A Starter Prompt Kit

AI #1 : The Interviewer. Use the prompt from Chapter 2, replacing the goal with your own.
AI #1 again : The Summarizer. Use the prompt from Chapter 3, in the same chat as the interviewer, once you have enough information.
AI #2 : The Doer. Open a new chat, paste the context brief, have the AI say what it understands first, and then paste the prompts one step at a time.

← All articles

Keep reading