Reverse-Engineering Prompt Creation: Let AI Read Your Ideal Text to Generate Instructions

Hello, I am Pokego, and I research prompts for writing high-quality articles with generative AI.

If you use AI to write text, you have likely felt:

"How should I write a prompt to get the text I want?"
"It's hard to put into words the specific style I want to convey to the AI."

Have you ever felt this kind of frustration?

The root cause of unsuccessful prompts is that the characteristics of your "personal writing style" often lie in areas that are difficult to fully articulate, even for yourself.

Habits in vocabulary, the rhythm of the narrative, how you lead to a conclusion, and so on...

No matter how detailed your instructions are, you cannot fully convey these "writer's unconscious habits."

However, if you reverse your thinking, a solution becomes clear.

Instead of telling the AI "write like this," you have it read your ideal text and ask it to "create instructions to reproduce this."

In this article, for those who want to be freed from the struggle of writing prompts from scratch, I cover:

• Basic steps to have AI read "correct" text and reverse-engineer a prompt

• Tips for improving the accuracy of reverse-engineered prompts through dialogue with AI

• Copyright considerations and precautions when using others' writing as source material

I discuss the above based on my own insights gained from extensive research and practice in prompt design.

Once you learn this reverse-engineering approach, the trial and error involved in prompt creation should decrease significantly.

Please use this as a reference and start by choosing one piece of ideal writing.

The real reason why writing prompts from scratch is "exhausting"

No matter what you write in a prompt, you get the feeling that the output is somehow "not your own writing."

The cause of this lies in a surprisingly simple place.

The problem is not how you write the prompt, but that you cannot verbalize the true nature of the "personal style" you want to convey.

Let's look at why this happens, step by step.

"Personal style" cannot be fully verbalized even by the author themselves

Writing contains many more "habits" than the author is aware of.

For example, the choice of vocabulary.

Some people prefer to use "but" instead of "however," while others might prefer to switch topics using "just."

Since these choices are made unconsciously when writing, even if you ask the author, "What are the characteristics of your writing?", they cannot put everything into words.

This is because elements like the rhythm of the sentences, how paragraphs are broken, and how the conclusion is reached are like "physical habits" naturally acquired through years of writing experience.

In other words, even if you write "write like me" in a prompt, all that gets across to the AI is a superficial style (such as polite vs. casual tone).

The deeper parts of your "personal style" remain unwritten in the prompt due to the limitations of verbalization.

The reason you feel like something is 'a bit off' even after giving detailed instructions in a prompt lies right here.

Because you cannot fully convey it in words, the trial and error process expands.

What happens when you can only write superficial style instructions in a prompt?

Every time you feel like it's 'a bit off,' you fall into a loop of editing, where you keep adding or rewriting instructions.

I have also repeatedly improved my prompts, but the more I obsessed over them, the more the instructions grew, and one prompt even reached 30,000 characters...😅

Of course, writing in detail improves accuracy.

However, the root problem is that you cannot verbalize your 'uniqueness' 100%, so even if you add more words, the feeling that 'something is still a little off' rarely disappears.

This cost of trial and error is the true nature of the 'difficulty' in writing prompts from scratch.

If so, let's change our perspective.

If you cannot fully verbalize it, stop trying to verbalize it and show the AI the 'real thing'.

This reverse approach is the 'reverse engineering' method I will introduce in the next chapter.

How to reverse engineer a prompt from your ideal text

If you cannot verbalize it, then don't.

So, how do you do it specifically?

Here, I will share the basic steps for having AI generate instructions by feeding it your ideal text, along with tips to further improve accuracy.

When you actually try it, you will find it is surprisingly simple.

Basic steps for having AI read an article and generate instructions

First, prepare one article you have written or a piece of text you want to use as a model.

It can be a note article, a blog post, or a business document.

Pass it to the AI and try asking it like this.

この文章と同じトーン・構成で新しい文章を書くための指示書（プロンプト）を作成してください。

Then, the AI will analyze the text and...

• the rhythm of the writing

• how paragraphs are structured

• how the reader is addressed

It will read these elements and propose a prompt to reproduce them using AI.

The AI will read the nuances you couldn't fully explain yourself from the actual text and summarize them into an instruction manual.

Instead of instructing it to 'write like this,' ask it to 'teach me how to reproduce this'.

This is reverse-engineering a prompt.

You might think, "Is that all?" but in reality, this idea of "asking the AI" is often a blind spot.

Since it is the AI itself that executes the prompt, it makes perfect sense to have the AI think of how to write instructions that it can easily understand.

I have also developed this approach by having the AI read my own article data to build a "writer profile," which I then use as a reference base for prompt design.

Related article:

Improving accuracy by specifying analytical perspectives in dialogue with AI

A prompt obtained by working backward will not be perfect on the first try.

This is to be expected, as the AI can only judge "what should be analyzed" in a general sense.

The key to further improving accuracy from there is to specify the analytical perspectives concretely from your side.

For example, the following additional instructions are effective:

• "Analyze the development patterns of the introduction section in more detail"

• "Focus on the tone of how the reader is addressed"

• "Include patterns for character count per paragraph and sentence length"

By narrowing down the perspectives in this way, the AI's analysis becomes deeper, and the accuracy of the prompt increases by another level.

I recommend approaching this with the mindset of "nurturing it through several exchanges" rather than trying to complete everything in a single conversation.

In my experience, you can often create a highly practical prompt in just two or three interactions.

Points to keep in mind for reverse-engineering prompts

Reverse engineering is a powerful time-saving technique, but there are some important points to keep in mind.

Here, I will introduce two points you should know before putting it into practice.

Copyright considerations when using someone else's writing as source material

There is no problem at all if you use your own writing as the "sample text" for reverse engineering.

What you need to be careful about is cases where you have the AI read someone else's writing.

For example, having the AI analyze an article by your favorite writer to create a prompt, and then writing your own article in that style.

This is acceptable if it remains just a "reference" for style, but if you go as far as reproducing the expressions and structure almost exactly, it could develop into a copyright issue.

As a basic rule, keep the following in mind:

• Using your own writing or public domain text is the safest approach

• When using someone else's writing as a reference, limit it to "analyzing stylistic trends" and use it as a guide for direction rather than copying expressions

• If using text shared by a client, check the contractual agreements

When in doubt, starting with your own past writing is the most reliable way.

Finish the parts that AI cannot fully replicate yourself

While reverse-engineered prompts are powerful, they are not a cure-all.

AI is good at reading the 'patterns' of text, but it cannot fully replicate the unique perspectives born from the writer's own experience and knowledge.

For example, primary information such as 'things the author noticed when they were running a web media site' is beyond the scope of what AI can extract from a sample text.

Therefore, once you receive the text generated by a reverse-engineered prompt, try checking the following two points:

Reverse engineering saves a significant amount of time in prompt creation, but the final polish is the writer's own job.

By being conscious of this 'fill it in yourself' step, you will be able to create high-quality text through co-creation with AI.

Summary: Prompts can be created simply by 'showing the correct answer first'

In this article, for those struggling with the effort of writing prompts from scratch, I have discussed:

• The real reason why writing prompts from scratch is 'exhausting'

• Practical steps to have AI read your ideal text and reverse-engineer a prompt

• Copyright considerations and precautions when using reverse-engineered prompts

I have talked about the above while incorporating the insights of the author, who has repeatedly researched and practiced prompt design.

The hardest part of creating a prompt is the task of putting your 'uniqueness' into words.

However,with a reverse-engineering approach, you can completely bypass the hurdle of putting things into words.

If you show the AI the "correct answer"—your ideal text—it will analyze the characteristics and turn them into a set of instructions.

Once you master this mechanism, the trial and error involved in prompt creation will be reduced significantly.

Start by choosing one piece of your past writing that you feel captures your personal style, and have the AI analyze it.

That small step will open up a new style of prompt creation for you!

✅ Poke-Go Content Here