Carlos Lastres
Carlos Lastres1,208 words

Japan's AI Problem Is Not Fear. 38.8% Just Cannot Find the Front Door.

The 2026 white paper on information and communications came out in July, and buried in the first chapter is a survey question I wish more people asked. Not why do you use this. Why don't you.

They put it to people in Japan, the United States, Germany and China who had never used a text generative AI service. In Japan, 42.4% said it is not needed for their life or work. 38.8% said they do not know how to use it.

38.8% is the highest of the four countries. The white paper says so directly.

The number that is not there

Look at what Japanese non users did not say.

Worried about data leaks, security or safety: 10.4% in Japan. 19.9% in the United States. 16.0% in Germany. 20.5% in China.

Worried about quality: 6.5% in Japan, against 13.4%, 14.2% and 26.0%.

Cannot change existing culture or habits: 2.7% in Japan. The lowest of the four by a distance.

This is the opposite of the story I hear at conferences. The received wisdom is that Japan is cautious, risk averse and slow to trust new technology, and that this is why adoption lags. The people who are actually not using it are the least worried group in the survey.

They are not afraid of it. They cannot find the way in.

What the numbers actually are

Overall usage experience in Japan came out at 58.8%, against 75.6% in the United States, 75.6% in Germany and 93.6% in China.

One honest footnote before anyone quotes the jump from last year's 26.7%. The 2025 survey added 15 to 19 year olds, who were not in previous rounds. On a comparable basis, counting only people aged 20 and over, Japan's figure is 52.2%. The rise is still large. It is not as large as the headline.

Teenagers are why. 91.7% of Japanese 15 to 19 year olds have used it. Then it falls: 78.2% in their twenties, 56.3% in their thirties, 49.0% in their forties, 44.2% in their fifties, 33.5% in their sixties.

Every country slopes down with age. What differs is the height and the steepness. Germany starts higher at 98.1% for teenagers and lands at 45.2% for people in their sixties. China barely slopes at all, 96.2% down to 86.5%. Japan starts near Germany and ends near a third.

A sixty point drop across the working population. In China it is ten.

Japan is using it as a dictionary

The survey asks a question I have not seen elsewhere. What is generative AI to you, as a thing?

In Japan the top answer is machine or tool at 38.3%. Second is dictionary or encyclopedia at 28.9%. Then secretary or assistant at 21.4%. Friend comes in at 10.9%.

The United States also leads with machine or tool at 46.0%, but its second answer is friend at 25.3%. Germany's second is counsellor at 41.0%. China's top answer is not machine at all, it is secretary or assistant at 44.7%, with friend at 39.9% and counsellor at 38.1%.

I keep coming back to the dictionary.

A dictionary is a thing you consult when you already know the precise question. You do not converse with it. You do not tell it what you are trying to do and let it ask you something back. You look up one word, get one answer, and close it.

That mental model does not fail loudly. It fails quietly, by producing mediocre results, which the user reasonably attributes to the tool. You ask a dictionary question, you get a dictionary answer, you conclude it is not that useful, and you stop.

Nobody in that story did anything wrong. The interface told them it was a search box, so they used it like a search box.

This is an onboarding failure and it is fixable

I build products. When 38.8% of the people who bounced off mine said they could not work out how to use it, I would not commission a study on cultural attitudes to technology. I would go and watch five people open it.

Here is what a Japanese person who has heard of ChatGPT sees on their first visit. An empty box. A cursor. Maybe a couple of example prompts in a language and register that do not sound like anything they would type.

That screen assumes you already know what the product is for. It works beautifully for the person who arrived with an intention and needs no instruction. It is close to useless for the person who was told this thing is important and went to look.

The tools are getting better at this and the Japanese localisation is generally fine. Fine translation is not the issue. The blank canvas is the issue, and it is a design decision, made by teams whose users all arrive with intent because those users are the ones who already adopted.

What I would do differently

If I were designing for the 38.8%, four things.

Start from a task, not a box. Not example prompts as decoration, but a short list of real jobs with a first step already taken. Summarise this document. Write a reply to this email. The gap between an example you can read and an example that is already running is enormous.

Show the conversation, not the query. The single most valuable thing a new user can see is a session where the model asks a clarifying question and the answer improves. That one demonstration kills the dictionary model faster than any explanation.

Make being wrong visible and cheap. The dictionary model persists partly because people assume a wrong answer means they broke it. Show what correcting looks like. Show the second attempt being better.

Design for the fifty year old first. The person under twenty five will find their way in regardless. Every design decision that helps someone who has never had a reason to use one of these tools also helps everyone else, and the reverse is not true.

Why this matters commercially

Japan has 58.8% individual usage against 93.6% in China, and 86.4% of Japanese companies now using generative AI somewhere in the business. That is a large and growing gap between what people do at work and what they do the rest of the time.

Companies here spend on training to close that gap. I think a good part of it is not a training problem. Training teaches people to operate an interface that failed to explain itself, which is an expensive way to patch a design defect.

And 38.8% is not a wall. It is the most encouraging number in the entire survey. Fear takes years to move. Not knowing where to start takes one good first screen.

Sources

← All posts