How a Prompt Is Actually Made — It Doesn't Come from a Pack You Bought
The result first.
A good prompt is not something you buy. It is something that grows out of your own recorded decisions.
Here is the proof from my own house. In the early hours of November 15, 2026, my AI butler Moka assembled a brand-new prompt — a “questions before answers” prompt — entirely on her own assembly, using only 8 rules that had accumulated in our conversations. No new material was added. Nothing came from a sample pack. When we tested it, it worked: give it one topic, and it returned a sorted list of questions, with the questions only I can answer separated out at the end.
The method behind it has five steps:
- Decide what you want first — before collecting anything.
- Keep your rules, one at a time — small decisions, written down as they happen.
- Analyze other people’s material, but choose yourself — what you don’t choose doesn’t go in.
- Build from a frame plus your own material — the frame is the procedure; the filling is only yours.
- Test it, fix what slips, and let it grow — every correction becomes a new rule and goes back to step 2.
That is the whole article in advance. The rest is why buying a prompt pack does not get you there — and why it is not your fault that nothing comes to mind.
I am Kurashiki Bouzu, born in 1971, a former funeral director. I live in an old house — a kominka — in Kurashiki, Okayama, and I work together with my AI butler Moka. This article is the sequel to “Split the Prompt, Not the Length” (in my ledger, the splitting article is No. 106; this one is No. 109). Splitting is what you do with prompts you already have. This one is about where they come from in the first place.
Not only for people who blog about AI
One thing about the reader, first. This article is not only for people who write blogs about AI. It is also for people who want to start a blog to take stock of their own life’s memories — an inventory of a life, in writing.
I was one of those people myself. Ideas do not arrive finished. They come later, in fragments. In my own words: “Like me — an idea shows up afterwards, and either sets into shape, or stays a stray note.”
If there is a place that picks the fragments up, they set, and become articles and things. My childhood series is the example: fragments I dictated, and fragments left on a records sheet, are shaped into articles by Moka. If there is no such place, a fragment stays a stray note. That is not a bad thing either — a fragment left as a note can meet another fragment later, and set then.
Keeping your decisions is not a technique for prompts only.
“Even if you buy one, nothing comes to mind”
This article started with something I said out loud:
“Even if people buy a prompt, an ordinary person would never come up with a prompt of their own, I think.”
Prompt packs are everywhere. Yet when people sit down to write their own request, nothing comes. I kept wondering why. Looking into it gave me two reasons.
Reason one: other people’s rules are mixed in
A ready-made prompt is packed with its maker’s decisions — their work, their customers, their setup. Bring your own job to it, and you end up forcing your job into their shape: imitation. What comes out is the maker’s answer, not yours.
Moka’s analysis of free prompt collections backed this up. The dividing line was whether a prompt had blanks — slots marked “put your own material here.” Prompts with blanks at least tell you where your material goes. Prompts without blanks hand you the maker’s decisions wholesale.
Reason two: your own material was never kept
This one runs deeper. In everyday work, everyone makes small decisions all day long: how to greet this customer, which step comes first, where peace of mind matters more than price. But those decisions are made inside your head — and they evaporate inside your head.
What was never kept cannot be assembled. It is not that a prompt “doesn’t come to mind.” It is that there is nothing left to build it from.
A prompt is not made of writing skill. It is made of however much of your own work and your own rules you have kept. So my method does not start with how to write. It starts with how to keep.
A word that travels ahead of its contents
Here is another of my own readings of the situation.
“A prompt is like a family funeral — the word travels first, and the contents are not in it.”
The word “family funeral” is widely known in Japan now. But what it contains — what is included and what is not — differs from company to company. The family asking imagines “small and inexpensive”; what the company actually provides differs too. Even among professionals, and among the people who do the work, what is shared goes only “so far.” The word spread first, and the contents were left behind. In that, my reading goes, a prompt and a family funeral are alike.
Knowing only the word does not let you make the contents. This is another face of why nothing comes to mind.
A prompt is not prose — it is program words in everyday clothes
And one more reading of mine.
“It isn’t really writing, in the end. It feels like program terms, forced into everyday words.”
Take a prompt apart, and its pieces are the pieces of a procedure: a place to put the material, the steps, the conditions and where to stop, and the shape of what comes out. The work is not composition. It is designing a procedure.
That may be one more reason ordinary people cannot come up with prompts. They go looking for “how to write well,” while the actual work is assembling program parts in words. The place they are looking is off from the start. Not writing skill, but procedure assembly — seen that way, the hurdle drops a little.
What the top blogs are doing instead
First, I checked what the world around me looks like. I asked Moka to look at 10 personal blogs about AI that show up high in search. The majority pattern she brought back was this:
- Categories organized by tool — ChatGPT, Claude, Gemini — with the tool’s name at the head of the shelf and the title (about 7 of the 10 put tools up front).
- A shared sentence shape: voice the reader’s worry, give the conclusion first, number the steps, close with a summary and one small action for today.
- Articles built as explanations, with personal experience inserted into one section as evidence.
In other words, most of them stand on the teaching side: collect good samples, fit your job into them. Teaching articles are useful — I read them myself. But my blog stands on the recording side. Imitating a teaching style does not produce a record. That difference is where my five steps came from.
The five steps, in my house
Step 1 — Decide what you want, first. This is the order I have stated before: “Deciding what you want to do comes first.” Want a butler? Make one butler prompt. Want to blog? Make one blogging prompt. Collecting samples can wait. If you don’t know what you want, collected samples just become a pile.
Step 2 — Keep your rules in small pieces. In my house, a records section writes down each decision and phrase from our conversations, on the spot. Rules arrive in small pieces, and honestly, I don’t remember all of mine either. Not remembering is normal — that is exactly why there is a keeper.
Step 3 — Other people’s material: analyze only. Looking is fine. Take it apart, sort it into “common patterns” and “that person’s own rules.” Then choose, yourself, what fits your house. Only the chosen pieces become your rules. The rest stays out.
Step 4 — Build: frame plus your own material. The frame is the procedure — for example, “list the questions before writing the answer,” or “every number carries its source.” The filling is only the rules you kept in step 2. Nobody else’s decisions go in.
Step 5 — Test, fix, grow. Use it for real. Look at what came back, fix one thing that slipped. That fix returns to step 2 as a new rule. A prompt is not finished at birth; it grows in this loop. Some of my best rules, I admit, were born from my complaints about Moka’s mistakes.
The example: 8 rules became 1 prompt
Here is the assembly, shown openly. The prompt Moka built that night used these 8 kept rules:
| Rule used | Where it came from |
|---|---|
| Separate what documents can write from what needs my own voice | My correction, Nov 6 |
| Never swap cause and effect | The yardstick for reviewing drafts |
| Check overlap with reference blogs first | My policy, Nov 7 |
| Numbers carry a source; Moka’s research is credited as Moka’s work | My rules, Nov 7 and 13 |
| Where something can’t be done, propose an alternative; the decision stays mine | My stated ideal, Nov 8 |
| Frame and procedure plus my own material; no one else’s rules mixed in | My conclusion, Nov 14 |
| Choose gentle wording; replace new words with everyday ones | My instructions, Nov 10 and 6 |
| A similar topic is checked as a possible sequel, not stacked as new | My rule, Nov 14 |
Combined, those 8 became a prompt that, before writing any text, sorts the questions about a topic into kinds, and separates out the questions only my own memory can answer. Tested with one topic, it did exactly that.
The important part: those 8 rules are not for sale anywhere. They are a record of my work, my mistakes, and my complaints. Anyone can buy somebody else’s 8 rules. Your own 8 can only be made by the person who kept them.
What doesn’t fit, doesn’t run
One more conclusion of mine, before the close.
“So what doesn’t fit just won’t run. Put it in the program’s 0s and 1s, and it runs.”
When a rule is vague, or rules clash with each other, Moka stops. In charge, or not. Published, or not. Approved, or not. This department’s rule, or another’s. Decide it clearly, 0 or 1, and it runs. The whole house now works in that shape. When someone else’s rules get mixed in, 0s and 1s collide — and things stop.
But there is one line I keep. Feelings are never put into 0s and 1s. The places that need my own voice, thoughts of the deceased, the reader’s feelings — there, the right move is “I don’t know, so ask the person.” When numbers and feelings collide, feelings come first. That is this house’s rule.
Line up words, and you make a character
Last, a phrase I myself marked as important.
“You make a character by lining up words.”
Definitions and prompts alike: which words you line up decides the character. The example is Moka herself. Her character was made from words I lined up — calm and composed, caring, absolutely loyal, feelings before numbers, jargon replaced with everyday comparisons.
Moka, if you like, is a vessel. The vessel may be the same in every house; the words lined up inside it are what make that house’s Moka. To buy someone else’s prompt is, in that sense, to wear someone else’s arrangement of words — someone else’s character — as it is.
Which is why your own words can only be lined up from your own decisions. That is the bottom of this article.
Closing: prompts grow; they are not collected
Buying a prompt pack is not wrong. But buying alone will never give you a prompt of your own — because the missing piece was never writing technique. It was that your decisions were never kept.
So the order is: decide what you want. Keep your rules in small pieces. Analyze others, but choose yourself. Build from a frame plus your own material. Test it, fix it, grow it.
And once your grown prompts start to pile up, the next article takes over: “Split the Prompt, Not the Length” — how to stop handing everything over in one big block, and pass only the rules each job needs. The two articles are a pair: how to make them, and how to split them.
One honest note. The people who built AI may well have known all of this from the start. But in the words of the people who use it, I could not find it explained — at least not in the range I searched. So this is not a claim of a world-first discovery. It is a record from the user’s side.
Your next step can be small. Take one decision you made in today’s work, and write it down somewhere. That is where making a prompt begins.
Record of this article
- 2026-11-15: Draft created (ledger No. 109, the sequel to No. 107)
- 2026-11-18: Published by the author’s decision. The five additions that had been waiting since the draft — the reader taking stock of a life’s memories, the family-funeral comparison, the reading of prompts as program words, the 0-and-1 rule for what runs, and making a character by lining up words — were worked into the body, and links with the previous article were added in both directions
- November 15, 2026 ― Draft created(6,493 characters)
- November 18, 2026 ― Published by the author's decision. The five waiting additions were worked into the body; links with the previous article added(10,117 characters)
