What Stops People Using AI Is Not the Same in Every Country
Before a brand builds a digital experience for Japan, it has a picture of the user it is building for. The picture is rarely written down. It shows up in what gets prioritized: how much reassurance to put on the page, how much explanation, how much friction to remove before someone will try the thing.
That picture is usually inherited rather than checked. So is much of what travels with a brand into a new market — we have written before about what happens to a set of brand guidelines that arrives without the reader it was written for. This one is different in a useful way: for at least one common version of it, there is public data.
Japan’s Ministry of Internal Affairs and Communications runs an annual survey of individuals on digital technology use, and publishes the results — including the underlying figures as downloadable data — in its Information and Communications White Paper. The 2026 edition covers Japan, the United States, Germany and China, asking the same questions in each.
The answers do not line up with one common version of that picture.
What the survey asked, and of whom
The 2026 white paper reports a survey of individuals conducted in the 2025 fiscal year, in four countries, on the use of generative AI services. The Japanese-language figures below are our translations of the published category labels; the numbers are as published.
Two things about the method matter before any of it is read.
The survey changed who it covers. The 2023 and 2024 rounds surveyed people aged 20 and over. The 2025 round added 15-to-19-year-olds, in all four countries. The white paper is explicit about this and supplies the like-for-like figure in a footnote:
In the 2023 and 2024 individual surveys the survey population was aged 20 and over, but the current 2025 individual survey newly adds ages 15 to 19. In the 2025 individual survey, the proportion aged 20 and over who answered that they use, or have used, some generative AI service was 52.2%.
So year-on-year growth in this data is not a clean comparison, and we are not going to use it as one. The cross-country comparisons below are all from the same 2025 round, where every country had the same change applied.
The second thing is how it was collected. It is an online panel survey, fielded in January and February 2026 through a research company’s monitor pool, drawn so that age and gender are not skewed. Valid responses ran to 1,236 in Japan and 624 in each of the other three countries. That is a real instrument and it is not a probability sample of a national population; the figures below should be read as one well-run survey rather than as a census.
One feature of that design is worth carrying into the tables. The sample was filled to an equal quota per age band — 206 per band in Japan, 104 per band elsewhere — with the youngest band capped at 15-to-19 and nobody aged 70 or over. Where a table below says “Japan,” it means a 15-to-69 sample built in equal age bands, and the four countries do not share an age structure for that design to sit on.
One more thing about who was asked. The same study ran a separate survey of companies — managers at firms of ten or more people that had already started digital work — and the two do not return the same picture. Asked which risks concern them about using generative AI, Japanese companies put internal information leakage highest of the four countries, at 46.4%.
Those are answers to a different question, from a different population. What follows is the individual survey. Both are worth knowing about, because an assumption about Japanese caution may have arrived from the corporate conversation rather than from anything users were asked.
The barrier people name is not distrust
Respondents who do not use text generative AI were asked why. The published figures, in percent:
| Reason given | Japan | US | Germany | China |
|---|---|---|---|---|
| No need for it in my life or work | 42.4 | 47.7 | 50.9 | 24.7 |
| I don’t know how to use it | 38.8 | 18.5 | 16.5 | 32.9 |
| No appealing service | 12.9 | 10.6 | 5.7 | 9.6 |
| Concern about leaks, security, safety | 10.4 | 19.9 | 16.0 | 20.5 |
| Concern about quality | 6.5 | 13.4 | 14.2 | 26.0 |
| Environment not in place | 6.5 | 11.6 | 9.0 | 23.3 |
| Too expensive | 5.0 | 6.9 | 7.5 | 20.5 |
| Cannot change existing culture or habits | 2.7 | 6.9 | 2.8 | 15.1 |
This table and the next one omit the published “other” category. Both allow multiple answers, so neither column adds to 100.
Read the second row and the fourth together.
Security is the reason Japanese respondents selected least often of the four countries — about half the American figure. Not knowing how to use it runs at 38.8 in Japan, against 18.5 in the United States and 16.5 in Germany.
This is not a story about Japan alone. China’s figure for not knowing how is 32.9, close behind Japan’s. Germany’s largest single reason, at 50.9, is simply not needing it. Every country’s profile is different, and the useful question is which profile the work has been designed against.
If the imported assumption is that a Japanese audience needs to be reassured before it will try something, this survey does not support it. What it points at instead is a gap in instruction.
Users are not especially sold on it
The same survey asked users why they use these services. Again in percent:
| Reason given | Japan | US | Germany | China |
|---|---|---|---|---|
| It is free or cheap | 53.9 | 43.0 | 50.8 | 35.6 |
| Saves time and effort; improves efficiency | 42.8 | 45.1 | 49.2 | 62.0 |
| Shows me varied ideas and perspectives | 34.1 | 35.8 | 39.0 | 50.5 |
| Turns my intentions or ideas into something | 25.9 | 34.1 | 27.5 | 48.5 |
| It organizes and suggests to suit my tastes and situation | 31.6 | 30.7 | 29.7 | 46.2 |
| I can ask things that are hard to ask people | 29.0 | 36.9 | 24.6 | 37.3 |
| The content seems accurate and neutral | 12.7 | 24.8 | 19.5 | 32.0 |
| To pass the time | 22.6 | 22.2 | 17.6 | 12.5 |
| Wanted to try a new technology | 11.1 | 21.0 | 14.8 | 27.4 |
| No particular reason | 7.2 | 4.2 | 4.0 | 0.3 |
Price heads the Japanese column, at 53.9. The efficiency answer — the one that reads like a business case — is selected less in Japan than in any of the other three.
And confidence in the output is lowest in Japan by a wide margin. Twelve point seven percent against China’s thirty-two. So the low security figure in the previous table is not a picture of a trusting audience. It is an audience that is not relying on the thing enough for trust to be the deciding question.
The last two rows run the same way. Japan is at 22.6 for passing the time against China’s 12.5, and at 7.2 for no particular reason at all against China’s 0.3. Small figures, pointing in the same direction as the rest of the column.
Frequency follows from that
The survey also asked how often users use these services.
| Japan | US | Germany | China | |
|---|---|---|---|---|
| Almost daily, 1 hour or more per day in total | 11.1 | 25.4 | 19.7 | 17.0 |
| Almost daily, under 1 hour per day in total | 18.8 | 22.5 | 25.8 | 25.5 |
| At least weekly (not daily) | 35.5 | 28.4 | 30.9 | 44.2 |
| At least monthly (not weekly) | 19.8 | 14.0 | 14.6 | 10.4 |
| At least yearly (not monthly) | 9.6 | 5.7 | 5.9 | 2.2 |
| Not in the past year | 5.1 | 4.0 | 3.0 | 0.7 |
Adding the two daily rows: about thirty percent in Japan, against roughly forty-eight in the United States and forty-five in Germany.
Taken as single answers, the weekly band is where both countries cluster — 35.5 in Japan and 28.4 in the United States. They separate once the two daily rows are combined, and in the heaviest band alone the distance is wide: 25.4% of American users spend over an hour a day, against 11.1% in Japan. Same tools, same year, and a different place in the week.
Where the gap shows up
The survey also asked, for six everyday situations, whether people reach for AI or for something else. The figures below add the two AI-led answers published for each situation — “AI only” and “mainly AI, other methods secondary” — and are our arithmetic on the published columns.
| Situation | Japan | US | Germany | China |
|---|---|---|---|---|
| Casually getting information on everyday questions or small talk (sport, entertainment, hobbies and so on) | 15.5 | 42.5 | 30.3 | 45.7 |
| Getting the outline of news and current events (politics, economy, society, disasters and so on) | 9.6 | 33.5 | 28.0 | 32.7 |
| Researching news and current events in depth (politics, economy, society, disasters and so on) | 10.3 | 31.8 | 30.1 | 35.4 |
| Seeking expert consultation about one’s mind, body, life or future (medicine, psychology, law, finance, career, family problems and so on) | 19.9 | 34.3 | 26.9 | 49.5 |
| Considering and deciding on everyday purchases (daily goods, appliances, clothing, food and so on) | 13.0 | 30.6 | 26.7 | 39.1 |
| Considering and deciding on purchases with long-term effects (housing, cars, insurance, education and so on) | 11.8 | 30.5 | 25.8 | 34.0 |
Japan sits below the other three in all six situations. This is not a purchase-specific effect — the distance is much the same whether the question is a football score or a mortgage.
Two rows are worth pausing on. On both purchase questions, the Japanese figure is under half the American one. For a brand whose Japanese customer research assumes an AI-assisted comparison step, that is the number to check the assumption against.
And the Japanese column peaks on the private situation — medicine, psychology, law, finance, career, family problems — at 19.9, the one place where the distance to Germany narrows to 7.0 points. It is the same shape as an earlier row: 29.0% gave “I can ask things that are hard to ask people” as a reason for using these services.
What that changes about what gets built
Take the three tables together and a shape appears that is worth designing against rather than around.
An audience that has largely tried the thing — the same survey puts Japanese usage experience at 58.8% across all ages, or 52.2% for the over-twenties — but uses it weekly rather than daily, does not rate its output highly, and, where it has stopped, says the obstacle was not knowing how.
That combination has a specific implication. The work is not in persuading someone that the thing is safe. It is in showing them what it is for on the occasion they have. Reassurance answers an objection that this survey does not find in large numbers. Instruction answers the one it does.
It also suggests where the risk of a wrong guess is concentrated. A launch built on the reassurance assumption would put its effort into trust signals, privacy language and third-party endorsement — all defensible, all standard, and none of them addressing the gap this survey actually shows. The work would pass every internal review and be aimed slightly to one side.
The limits of what this shows
The honest boundary matters more here than usual, because the figures above are precise and precision is persuasive.
It reports on generative AI services, and it covers the moment of deciding — including deciding what to buy. What it does not reach is the interface where a decision gets carried out: a checkout, a booking flow, an onboarding screen, or any particular product. The gap between “this is where a national sample sits on one category of tool” and “this is what my users will do” is not a gap statistics can close.
It is one survey, from one ministry, in one year. Its own footnote shows the population definition moving between rounds.
And it describes a distribution, not a person. Roughly a fifth of Japanese respondents who do not use these services gave answers about appeal and quality rather than instruction. Building only for the heaviest row in a table is its own kind of error.
What the data is good for is narrower and more useful: it shows that an inherited assumption is checkable, and that at least one common version of it does not survive contact with a public dataset. That is worth knowing before a budget is committed rather than after, and it is the kind of thing a small piloted build with real users is for — the questions this survey raises are answerable about your own product, and only about your own product.
Source: 令和8年版 情報通信白書 / Information and Communications in Japan: White Paper 2026, Ministry of Internal Affairs and Communications, published 24 July 2026. Figures from the published dataset (Ⅰ-1-1-4, Ⅰ-1-1-5, Ⅰ-1-1-6, Ⅰ-1-1-8, Ⅰ-1-1-12), which the Ministry attributes to its 2026 study “Research on trends in the research and development of the latest information and communications technology and in digital utilization, in Japan and overseas.” Survey method and response counts from the white paper’s appendix (付注). https://www.soumu.go.jp/johotsusintokei/whitepaper/ja/r08/html/datashu.html — all translations from the Japanese are our own.