What Humans Decide in AI-Assisted Article Creation | Netarie's Production Policy
In this article, using the question 'Was this article written by AI?' as a starting point, I will reflect on how AI is used in Netarie's article production from the perspective of Tsuyoshi Watataru. Rather than hiding the use of AI or emphasizing only its convenience, I have compiled a production policy aimed at refining text into something the client is satisfied with publishing, without arbitrarily supplementing their experiences or thoughts.
'Was this article written by AI?'
At first glance, it seems like a question that could be answered with a yes or no.
However, in reality, it is a bit difficult.
Transcribing audio. Categorizing long notes. Comparing headings. Finding areas where explanations are insufficient. Checking for typos or repetitions.
If AI is used at any of these points, should the entire article be called 'AI-written'? Conversely, if a human makes final adjustments, does that mean it was entirely written by a human? That also feels a bit off.
Hello. I am Tsuyoshi Watataru.
Under the trade name Netarie, I provide support for small business owners to publish on note.
I gather material such as notes, audio, past articles, and daily insights from within their work, and refine them into note articles that are searchable and convey their personality.
In this piece, I will write about where I use AI in article production, where human judgment begins, and what I verify to ensure the client's own words are not erased.
I hope this serves as a reference for those considering hiring me for article production.
The phrase 'used AI' does not explain how an article is made
There are many steps before an article is published.
First, we decide who the target audience is and what to write.
We collect notes and experiences that might be useful.
From those, we select the content to keep for the current article.
We create an outline, refine the text, and research necessary information.
Finally, we confirm whether it is acceptable to publish under the client's name.
AI can be involved in various parts of this process.
Therefore, the single phrase 'used AI' does not explain how the article was actually made.
Even if the majority of the text is generated by AI, the individual's experience and judgment may still remain. Conversely, even if a person types every sentence, it can still end up being nothing but generalities heard somewhere else.
What matters is not whether it was written by AI or by a human, nor the ratio of the two.
Whose material did it start from?
Who chose the content to remain?
Who made the final decision to publish?
I believe these are far more important, both when reading articles and when creating them.
The more you polish text for readability using AI, the more the individual's own voice can disappear
I think what people who outsource article creation are worried about is not just the existence of AI as a tool.
It is the so-called side effects of AI.
Will things I didn't say be written?
Will small achievements be turned into overly grand success stories?
Will content I spoke about with hesitation be stated as a strong conclusion?
Even if AI makes the text easier to read, you might feel, "This isn't my writing."
Sometimes the sentence endings or word choices are different, but in many cases, the "AI feel" becomes strong because the places where the person paused are processed away too cleanly.
For example, suppose you had an experience where you didn't recommend a service to a client.
If you have AI write the text, it will usually come back like this:
Making proposals that are truly necessary for the client leads to long-term trust.
The content is not wrong.
However, what the person wanted to leave behind might have been something more unpolished.
I wanted the sale.
So, honestly, I also had the urge to recommend it.
But as I listened to them, I thought it wasn't necessary for this person right now.
After much hesitation, I didn't recommend it this time.
Something like that.
A person's work is reflected in what they saw and why they decided not to sell. To make an article sound like the person, it is not enough to just add their habitual phrases.
The parts where they hesitated a little. The reasons they couldn't easily state something. The things they were honestly conflicted about.
It is necessary to leave behind such traces of judgment.
At Netalie, article production begins with raw material, not a finished draft.
Even when using AI, I believe raw material is essential to retain a human touch. If the material consists of primary information, the resulting text will carry the person's unique scent, even when AI is used.
When I work, I receive items like the following from my clients.
Bullet-point notes, audio recorded on a smartphone, replies to inquiries, past articles, and sentences jotted down during work.
None of these might be ready to become an article as they are.
However, it is precisely because the material is unpolished that certain words remain.
Explanations repeated multiple times.
Stories that were so obvious to the person themselves that they never thought to turn them into an article.
Parts where they slightly rephrased something while speaking.
Judgments about how they used to think versus how they think now.
While looking at such parts, I search for the material that will become the core of the article.
I am not creating a text that sounds like the person from nothing.
I pick up what is already present in the person's work and rearrange it into an order that conveys it to the reader.
I make full use of AI during that process.
Dividing long audio recordings by topic.
Finding parts in multiple documents where the same theme is discussed.
Checking if the content overlaps with past articles.
Creating multiple article structures and comparing the differences.
Finding explanations that are missing for first-time readers.
Checking if the same content is being rephrased repeatedly.
I also use AI when creating the draft.
However, I do not hand over the text AI initially produces as a finished manuscript as is.
I do not use AI to skip the verification process.
This is to broaden the scope of comparing structures and expressions, including materials that are easy to overlook.
The job that humans are responsible for is determining the meaning of the materials.
AI can find relevant parts within the materials.
It can also suggest headings. It can shorten sentences or supplement explanations.
However, it cannot decide whether the materials found should be kept in the article.
Even for the same topic, how it is handled changes depending on the purpose of the article.
Is it an article explaining things to someone visiting for the first time via search?
Is it an article conveying a way of thinking to someone who already reads your posts?
Is it an article answering the concerns of someone considering a service?
There are also topics that are better left out of the current article and separated into a different one.
There is also content that should not be published because it could identify individuals or clients. In article production, before writing the text, we decide what this article is responsible for.
Also, when using AI to organize customer information or internal company information, we do not input it as is; we handle it only after removing personal information and confidential data that is unnecessary for article production.
Based on that, humans verify the following points:
Which experience should be the focus?
Which parts should be cut?
How specifically should it be written?
Have we added things the person did not say?
Are we stirring up anxiety more than necessary?
Are we inflating achievements to make the service look better?
Are we stating things strongly where the person spoke cautiously?
AI can provide candidates.
But it does not decide what to keep or how much to say. I believe this is the significance of human involvement in article production.
AI is also used for organizing materials, structuring, and drafting. However, humans decide what to keep, how much to say, and whether to publish it under the person's name.
What is a manuscript that can be published under one's own name in Natalier's AI article production?
What should remain in AI-assisted article production is not the habit of polished writing, but the traces of what the person saw, where they hesitated, and what they decided to keep.
AI sometimes supplements explanations to make them sound natural as text.
Therefore, once a draft is complete, we check not only for readability but also whether it has strayed from the original materials.
Has any story that the person did not experience been added?
Are the achievements or figures larger than they actually are?
Has content that was spoken with hesitation been turned into a strong assertion?
When dealing with systems, fees, product features, etc., I do not use AI responses as the sole basis; I also verify them against original documents and official information.
However, there are things that I, as the creator, cannot judge on my own.
Was that event really perceived that way?
Is there any problem with publishing it with this phrasing?
Is it too assertive? Conversely, is it too weak?
Finally, I return the manuscript to the person for their review.
What I look for is not just typos or the quality of the writing.
It is about whether 'what they said remains'
and 'whether they can publish it as their own thoughts'.
These are the points.
Erasing the 'AI-ness' is not the ultimate goal. The goal is to ensure the manuscript is something the person can put their name on. I confirm up to that point to complete the article.
If it starts with the person's own material, retains traces of the person's judgment, and is finally confirmed by the person's own eyes, I believe it is an article that can be published under their name, even if it involves AI-assisted creation or outsourced production.
For those who want to start by preserving the material within their work
Before refining text with AI, it is necessary to preserve the source material.
For those who want to preserve the questions they are repeatedly asked by clients, I have compiled a guide on how to create a question ledger.
→ How to create a 'Question Ledger' to turn client questions into articles
If you want to preserve the reasons you thought of on the spot or specific situations, there is also a method using audio.
→ How to preserve article material using voice input
For those who want to see the whole picture of how to turn the material in their work into articles—not just how to use AI, but also using memos, questions, and audio—look here.
→ Article ideas are in your work | How to turn memos, questions, and audio into note articles
A list of services and the latest information are compiled on the official website. → Netarie Official Website
[About Production] This article is based on the structure, notes, and research of Tsuyoshi Watataru, using AI as an assistant for drafting and organizing, with final verification and editing performed by the author himself. The angle of the article, what content to keep, what to cut, and the decision to publish are all determined by Tsuyoshi Watataru.