Writing Was Too Early for Humanity - Humans Who Don't Read, AI That Does -

"Kenshi Yonezu 260 yen"

In February 2025, a certain post became a hot topic on social media.

This text, featuring an artist's name and a price, side by side. Do you understand what it represents?

In fact, this is an actual notice posted at a certain train station.

It was written to inform passengers that the fare from Hakata Station to the nearest station, Tojinmachi Station, was 260 yen when Kenshi Yonezu's concert was held at the PayPay Dome. This post spread widely, "Kenshi Yonezu 260 yen" trended on social media, and it was even picked up by online news.

The trigger for its spread was likely the impact of seeing something that looked as if Kenshi Yonezu could be bought for 260 yen, as well as the uniqueness of the phrasing. However, my impression the moment I first saw this post was quite different.

"How concise, easy to understand, and well-organized this text is," I was deeply impressed. I even thought it was beautiful.

What I felt was beautiful was not the words themselves. I felt it was beautiful to see these words functioning and solving a problem precisely.

The business impact of this notice is immense. The cost of station staff having to respond to passengers asking questions one by one must have been significantly reduced. That alone would be more than enough as a problem-solving measure, but because it spread on social media, the impact became even greater. It is rare to find a case where a problem solved through words has had such a positive influence on business.

Writing Harassment

Here, I would like to introduce another case regarding station notices that spread on social media.

On the purchase screen of an automatic ticket machine, the following label was affixed.

This post also spread widely on social media. However, in stark contrast to the situation where "Kenshi Yonezu 260 yen" was highly praised, the reaction on social media to this label was disastrous. The instructions on the automatic ticket machine were likely difficult to understand, leading many people to ask the station staff, so this label was probably posted to prevent that. However, the fundamental problem is that the instructions on the automatic ticket machine are difficult to understand. Writing difficult-to-understand text and then saying "Please read the text" can no longer be called anything but harassment. It is writing harassment.

Both are notices at a train station. Nevertheless, their evaluations are the exact opposite. Where on earth does this difference come from?

It is whether or not it is written on the premise that "humans do not read text".

Insights from a Writing Skills Workshop

Let me change the subject a little and talk about my own experience.

I work as a writer, and based on that experience, I serve as an instructor for writing skills workshops and in-house seminars. There, I assign a preliminary task. If you are reading this note now and have the time, I would like you to stop scrolling for a moment and actually try the task. By actually doing it, your understanding of this note should improve significantly.

For example, I give out an assignment like this.

That is all there is to it. I hope you will try tackling this assignment.

Have you tried it? Now, let's move on.

When I first started as a course instructor, I thought the purpose of the assignment was to test the writing ability of the students. That is only natural for a writing course. However, as the sessions went on, I realized a significant fact.

Many of the students were answering as if they hadn't read the assignment at all.

Since I am also a UX (user experience) expert, I design this assignment so that it allows students to experience as much of my practical work as possible. Furthermore, because I design it in a form very close to actual work, as if it were a request from a boss or client, I aim for it to be something that can be immediately applied to daily tasks. Therefore, if one reads the assignment carefully and acts accordingly, they will reach the answer I am looking for (the output that bosses and clients look for in practice) to a significant degree. Naturally, there are students who read the assignment well, complete the instructed tasks, and arrive at excellent answers. However, that number is very small every time.

Let me explain specifically. First, the beginning part.

There are always students who read only this and submit a catchphrase. For example, in the case of this assignment, it looks like this.

Since the assignment is for "copy that increases the registration rate when placed above the registration button," it is not a catchphrase. Nevertheless, because they have not read the assignment carefully, they likely misinterpret it as being asked for a catchphrase and submit one.

And the following part is the same.

Just by doing what is written here, the probability of reaching a good answer increases. Nevertheless, because theyskip over this part, they end up providing an answer different from what I was looking for.

Before I give feedback on this preliminary assignment, I explain that "humans do not read text." And when this assignment is presented again during the feedback session, some studentsrealize firsthand that they were the ones who hadn't read the text.

From these experiences, I have come to realize once again that "humans are creatures that do not read text." And I realized that before the ability to write, the ability to read text is necessary first.

Reading this far, some might think, "The way you write the assignment is bad!" or "What kind of UX expert are you!"

If I were in the same position, I would definitely have thought that way too. However, although it is very difficult to convey, there actually exists a dream-like, perfect partner who accurately grasps the assignments I give, completes the tasks exactly as instructed, and proposes the output I expect.

That is generative AI.

Here is the result of inputting the exact same text as the assignment above into Google Gemini.

In fact, copy like the one in this assignment is called "microcopy," and it contributes significantly to improving business metrics. I work on microcopy at tech companies including Rakuten, and even from my perspective, all of these are expected to improve registration rates and are fully capable of withstanding practical use.

In particular, the text above is based on principles of psychology and behavioral economics, such as clear benefits (try it with a free trial), simple procedures (easy 1-minute registration), and anxiety relief (cancel anytime), and it can be said that the results of the research are clearly reflected.

Furthermore, the following part of the assignment:

My intention in including this sentence was to make them realize that since long copy placed above a button won't be read, short copy is optimal. Naturally, there are many people who didn't read this or didn't grasp the intent even if they did, and some students submit copy that is extremely long to be placed above the button.

However, the AI returned an accurate answer here as well. This is the following part from Google Gemini.

It perfectly grasped my intent as the requester.

By the way, just to be clear, this assignment was not adjusted to match the AI's output. As I have continued to lecture for several years, I thought about how students could arrive at copy that solves business problems in a way that they can reproduce and work independently in future practical work, not limited to assignments in lectures. The result of those improvements is the "things you must do before thinking" mentioned above. Later, since generative AI became widespread, I tried inputting it as is, and it returned the results I expected.how students can arrive at copy that solves business problems in a way that they can reproduce and work independently in future practical workThe result of those improvements is the "things you must do before thinking" mentioned above. Later, since generative AI became widespread, I tried inputting it as is, and it returned the results I expected.

From this case, we can see that AI can output excellent copy. And most importantly, AI reads all the given text and can act according to the text.Unlike humans, AI does not misunderstand the meaning of text or skip over it.

The importance of writing skills increases in the AI era

Humans do not read text. Therefore, no matter how hard you try to write text, there was an overwhelming inefficiency in that it wouldn't be read. Furthermore, even if it is read, there is no guarantee that it will be understood as intended. In other words, the act of conveying text written by humans to humans had two walls:the "wall of completion" and the "wall of comprehension."

For example, if you wrote text with 100 units of information for humans, and 50% read it and 25% understood it correctly, then 75% would have been wasted. It looks miserable when put into numbers, but try to imagine the text you write for business every day. Wouldn't it be considered good if 25% was conveyed?

However, AI is different. First of all, AI will definitely read all the text you write.For a human who writes text daily, there is nothing more grateful than this.Furthermore, as long as you can write logical and accurate text, AI will output the results you expect.

It reads 100% of what you give it and understands 100% of it. Of course, if the input text or content becomes complex, the AI will naturally skip parts of the text or misinterpret things. However, even taking that into account, for AI, text is a means of conveying information far more efficiently than it is for humans.

The automatic ticket machine at the station mentioned above had a label on it that said, "Please read the text," but if this were for an AI rather than a human, such a label would not have been necessary. This is because an AI would read all the text and complete the task of buying a ticket without any issues (assuming, of course, that the text on the ticket machine was written accurately).

Because of this overwhelming advantage, the importance of writing skills will likely increase as AI becomes more widespread. And what is required there is not poetic, beautiful, emotional writing, but logical writing that conveys things accurately and clearly.

How to write on the premise that "humans do not read text"

At the beginning, I compared two cases and stated that the difference lies in whether or not the premise is that "humans do not read text."

So, when we convey things through text to humans who do not read, what should we be careful about? One of them is reducing the number of characters to lower the cognitive load on the brain.

For example, manga artist Yusei Matsui says the following in "Jump Manga School," a creative course for manga artists.

By reducing even one character without changing what you want to convey, readability is dramatically improved.

For example, suppose there is the following text.

In this text, what happens if you push the task of "reducing the character count by even one character" to the extreme? Here it is.

I hope you can understand how efficient and excellent this text communication is.

Writing was too early for humanity

So why, in the first place, can't humans read text?

First of all, human hardware has not been updated since the hunter-gatherer era 200,000 years ago.

John J. Ratey, an associate clinical professor of psychiatry at Harvard Medical School, states the following:

On the other hand, it is said that the technology known as "writing" was born 5,000 years ago.

In the staggering 200,000-year history of humanity, only 5,000 years have passed since writing began to be used. In other words, the human hardware has not had the software installed to process the technology of writing.

The state of using the recently born technology of writing with hardware that hasn't been updated in 200,000 years could almost be called a bug. Naturally, glitches will occur. The tool known as "text," which is a more complex configuration of those characters, was too early for humanity.

Even for me, whose job is to write, I have always thought that text is such an inconvenient tool (of course, I have received great benefits from its convenience). I have always wondered, what is this time spent going back and forth on Slack when it could be solved in 5 seconds by talking?

Text is inherently an inconvenient tool for humans. Therefore, even if you are not good at reading or writing, it is, in a sense, natural.

Commonalities with the world's oldest writing

Finally, I would like to introduce a very interesting case.

According to Yuval Noah Harari's 'Sapiens: A Brief History of Humankind', about 5,000 years ago, our ancestors left behind the following text to convey the message: "A total of 29,086 units of barley were received over 37 months. Kushim (signature)."

Doesn't this list of nouns and numbers, said to be the "world's oldest writing," look familiar?

Yes, it has the exact same structure as the "Kenshi Yonezu 260 yen" introduced at the beginning.

I think you can understand how optimal text is for human hardware. And the fact that humanity has not been updated for at least 5,000 years seems to be certain.