I think you’ve opened an article like “100 Best Prompts” at least once.
You copy and paste it to try it out, and then tilt your head a little. It doesn’t return the kind of answer that was introduced. On the other hand, even though they are using the same AI, some people are getting incredibly accurate answers. Where does this difference come from? That is the starting point for today.
I’ll write the conclusion first. What affects the quality of an AI’s answer the most is not the cleverness of the phrasing, but how accurately you were able to explain your situation. “Explanation” over “spells” is my argument for today.
By reading this article, you will gain the following three things.
-
Drawing the line between where tricks work and where they don’t
-
Four elements to keep in mind when writing instructions
-
A 5-minute “explanation test”
As the first installment of this series, I’d like to start with the foundation. From the next installment onwards, we will move on to how to use this for brainstorming and consulting on entrepreneurship.
[Image: An image contrasting a “person casting a spell” and a “person explaining the situation”]
🪄 Where tricks work, and where they don’t
To avoid any misunderstanding, I’ll write this first: it’s not that all prompt ingenuity is meaningless. Specifying formats like “output in a table” or “within 300 characters,” conveying a role in one sentence, showing examples of the desired output, and organizing information with headings before passing it on—these kinds of efforts really do work.
However, these are tools to make explanations easier to convey, and they are not a substitute for an explanation. I want to think about these separately.
-
Works as a tool: Specifying format, one-sentence role, examples, structuring information
-
Does not work as a substitute for explanation: Polishing only the phrasing without conveying your own situation, repeating strong words like “absolutely” or “must,” and copying and pasting templates made for someone else’s situation
To use an analogy, the tools are the “wrapping paper,” and the explanation is the “contents.” If the wrapping paper is beautiful, it’s easy to hand over. But if the box is empty, the recipient will be troubled. What determines the content of the answer is the material the AI thinks with—in other words, how much of your situation you were able to convey.
🧑💼 AI is a “highly capable newcomer who knows nothing about your circumstances”
I’d like to provide one metaphor that is easy to visualize.
Imagine a very capable freelance consultant who started working on their first day. They have a wide range of knowledge and are quick-witted. But they don’t know anything about your customers, your budget, or your past failures yet.
What happens if you just say to that person, “Please give me some side business ideas”? What comes back will be generalities that apply to anyone. That’s not a problem of capability, but because the information you provided was just one line, right?
AI is the same; it basically doesn’t know the circumstances not written in the conversation. It will try to fill in the missing parts with guesses, but those guesses tend to assume an “average someone.”
🔍 Let’s compare the same question with and without an explanation
Let’s look at a concrete example.
One-line version
副業で何か始めたい。おすすめは？
Version with added explanation
【目的】3か月以内に、月1万円の売上が出る最初の副業を決めたい
【状況】平日夜に2時間、週末に半日使える。初期費用は3万円まで。
人前で話すのは苦手。文章を書くのは苦にならない
【基準】在庫を持たない／1人で完結する／2週間で「売れそうか」を検証できる
【材料】以前ハンドメイド販売を試したが、制作時間が取れず1か月でやめた
【お願い】候補を3つ。各候補について、最初の2週間でやる検証方法まで
The latter is longer. But the material the AI can use is completely different. You can narrow down candidates based on time constraints, exclude proposals with the same structure as past failures, and compare them using a “verifiability” yardstick.
The output format was just slightly added to the final [request]. The point of this comparison is that the amount of explanation determines the specificity of the answer, that is the point of this comparison.
🧭 Four elements for writing instructions
To avoid thinking from scratch every time, keep it in a template.
1. Purpose: What is the consultation for, and what determines success? Not “I want ideas,” but “I want to choose one side hustle that earns 10,000 yen a month in 3 months.”
2. Situation: Premises and constraints. Available time, money, skills, and things you don’t want to do. This is the part most easily omitted, and the most effective.
3. Criteria: Conditions for a good answer. “Safety-oriented or speed-oriented?” “Novelty or certainty?” If you don’t state the criteria, the AI will choose the safe middle ground.
4. Materials: Things you’ve already tried, numbers you have, and the history of what didn’t work. Failure stories are actually the most valuable material.
Output format (making it a table, character limits, etc.) can just be added at the end. Small techniques are just the finishing touches.
🔄 Paradox: The better you are at explaining, the better you can use AI
I want to write one slightly paradoxical thing here.
Everyone thinks that the more you polish your prompt skills, the better you get. But in my view, a different ability is what really counts. The ability to put your own situation into words that others can understand. It’s closer to the ability to organize your thoughts than writing ability.
Moreover, I believe this gap will tend to widen as AI becomes smarter. The reason is simple: while the model will absorb differences in the fine phrasing of instructions, only the person themselves knows “what they are looking for,” no matter how smart the AI gets.
Recently, there has been a movement to call the entire design of information passed to AI “context engineering” and emphasize it. Even if the wording changes, the point is the same, and I think it means what you provide is what wins.
🧪 A 5-minute “explanation test”
Rather than theory, it’s faster to try it once. Here are the steps you can take starting today.
-
Decide on one question you want to ask the AI now
-
First, write it in one line and send it as is
-
Open a different chat and send a version that includes four elements (purpose, situation, criteria, and materials)
-
Compare the two answers and see which one reflects your situation better
-
The parts where there is a difference are the pieces of information you usually omit in your explanations
Point 5 is particularly interesting. The differences become a list of the “assumptions you unconsciously skip.” You should be able to see the items you need to write first next time.
💡 Perspective ── Tricks are consumables, explanations are assets
This is the point I most want to convey.
Spells of the “adding this one sentence improves accuracy” variety change in effectiveness when the model is updated. They are like consumables with an expiration date.
On the other hand, being able to properly explain your business or situation is something you can bring to any AI. Moreover, once you have articulated it, you can reuse it in the future. I think that cultivating explanations as assets is more effective in the long run than learning tricks.
✨ Summary
Techniques like specifying formats or providing examples are effective as tools to make explanations easier to understand. However, it was the “explanation”—the purpose, situation, criteria, and materials—that determined the content of the answer. Think of the AI as a talented new hire who doesn’t know your circumstances, and provide the situation first. Then, add a little bit of formatting on top of that. I believe that just by following this order, the resolution of the answers you receive will change.
Next time, I will apply this way of thinking to “brainstorming for startups” and specifically organize what needs to be explained for the AI to function as a sounding board.
Out of the four elements, how many were you able to write in the questions you recently threw at the AI? If you look back, you might find your own habits in the items you omitted 🎯
