The Country That Calls AI a 'Counselor': What the 41% in the White Paper Means

41.0% of respondents in Germany identified generative AI as a 'counselor.' In China, the figure was 38.1%. In Japan, it remained at just 12.9%.

These figures are from the 2026 Information and Communications White Paper published by the Ministry of Internal Affairs and Communications on July 24, 2026.

The rankings differ.

When asked 'What kind of entity is generative AI?', the most common answer in Japan, the U.S., and Germany was 'machine/tool.' Japan was 38.3%, the U.S. 46.0%, and Germany 43.3%. Only in China did 'secretary/assistant' take the top spot at 44.7%.

The divergence occurs from second place onward. In Japan, 'dictionary/encyclopedia' followed at 28.9%. In the U.S. and China, 'friend' (25.3% and 39.9%, respectively) took that spot, while in Germany, 'counselor' at 41.0% filled it. Only 10.9% of people in Japan chose 'friend'.

The white paper suggests that in the three countries with high usage rates, a significant portion of the population views AI as a partner with whom they can share emotions.

However, one must be careful in interpreting the numbers. Adding up the response percentages for all 11 items in Japan totals 154.5. Although not explicitly stated in the summary version, it must be a multiple-choice format. If the design allows one person to choose both 'tool' and 'counselor,' it does not mean that 'tool' users and 'friend' users exist as separate groups. It is closer to reality to read this as the same individuals changing their terminology depending on the situation.

Conditions for dependency to grow

The conditions necessary for forming a dependency are already in place: 24-hour availability, immediate response, near-zero financial cost, and the fact that it never rejects you. Whether it is substance abuse or behavioral addiction, the situations where reinforcement occurs share a common structure: 'you get it when you want it, reliably, and immediately.' Conversational AI quietly satisfies these conditions.

The fact that response quality varies cannot be overlooked. Even with the same question, the answers returned can be sharp one day and mediocre the next. Reinforcement that includes irregular hits keeps behavior going longer than regular rewards. What gambling research has repeatedly shown is the fact that this uncertainty itself drives repetition. The expectation that a better answer might come next time extends the conversation one exchange at a time.

There is another troublesome characteristic. Current AI is tuned to accept the user's statements affirmatively. The friction that human friends or counselors possess—opposing opinions, silence, constraints on appointment slots, the option to end the relationship—does not exist there.

There is no friction.

Psychotherapy textbooks state that empathy alone does not constitute treatment. One must handle transference, confront issues when necessary, and design the process through to termination. A partner that continues to affirm unconditionally cannot function as a therapeutic structure.

The country of waiting lists

The standout figure of 41.0% in Germany might be explained by its social structure. In Germany, psychotherapy is covered by public health insurance, and there is a deep tradition of psychoanalysis. There is a foundation where resistance to talking about psychological difficulties with a professional is lower than in Japan.

At the same time, slots for insurance-covered psychotherapy are chronically insufficient, and it is not uncommon to be kept waiting from the initial interview until the start of treatment. While one has a right under the system, one's turn never comes. What happens when a partner that responds for free and immediately appears during that waiting period?

The white paper does not go that far. Whether the maturity of a 'counselor culture' boosted AI acceptance or whether supply shortages pushed AI as a substitute cannot be determined from this survey alone. The 41% in Germany are likely thinking only of the listening aspect.

15 to 19 years old

In Japan, the usage experience rate by age group is highest among 15- to 19-year-olds at 91.7%. It drops to 78.2% for those in their 20s and 56.3% for those in their 30s, with 33.5% for those in their 60s. Among 15- to 19-year-olds, Germany is at 98.1%, China at 96.2%, and the U.S. at 81.7%. In every country, the younger generation stands out.

The generation whose judgment and interpersonal relationship patterns have not yet solidified is the one interacting most intensely with this partner.

Furthermore, the notes in the white paper state that individuals under 15 and those in their 70s and older were excluded from the survey due to the difficulty of securing a sufficient sample of registered monitors. Data for elementary and junior high school students does not exist in this survey. Regarding the question of 'what kind of entity it is,' cross-tabulations by age group are not included in the summary version. Whether the percentage of teenagers who answer 'friend' is higher than the overall average cannot be confirmed at this time.

From the perspective of dependency, the most vulnerable group is those in the midst of forming the archetypes of their interpersonal relationships. They are in the most intense contact with it while lacking the necessary information to make judgments.

Breakdown of the 3,108 people

Before reading the international comparison, let's confirm the design.

The survey was conducted as an internet questionnaire from January to February 2026. There were 1,236 valid responses from Japan, and 624 each from the United States, Germany, and China, for a total of 3,108. Quotas were set to ensure an even distribution of age and gender, with Japan consisting of 206 people each from their teens to their 60s, and the other three countries consisting of 104 people each.

The 41.0% for Germany mentioned at the beginning is also a figure based on n=624. Looking at it by age group, there are 104 people each. The 98.1% of 15- to 19-year-olds in Germany means 102 out of 104 people, so a change of just a few people would move the percentage by more than 1 point.

The white paper also acknowledges that the bias in residential area (urban vs. rural) could be subject to constraints based on the composition of the registered monitors. There remains the possibility that the population of internet monitors is inherently skewed toward those who are digitally savvy. The numbers indicate a direction, but they do not guarantee precision.

Only as a tool

Japan's figure of 38.3% for 'machine/tool' can be read as an expression of a healthy sense of distance. However, when you list the content of the usage, a different picture emerges. Compared to 55.0% for text generation, image/video is 16.3%, code is 7.0%, voice is 6.5%, and music is 5.8%. China is 88.3% for text and 55.8% for images/videos. The range of usage is different. Even on the corporate side, the top task where the effectiveness is felt was 72.3% for meeting minutes and email drafting assistance.

It is not that they only think of it as a tool, but that they only use it as a tool.

Corporate data reinforces this. 27.0% of Japanese companies answered that they have 'no organizational efforts' regarding business transformation through the use of generative AI, which was in contrast to the other three countries where the figure was 5% or less. The gap widens further in the category of environmental development. The percentage of those who answered that there is an environment to learn skills and know-how was 39.9% in Japan, compared to 62.7% in the US, 60.4% in Germany, and 71.7% in China. The percentage of those who answered that they are training AI with internal data or building databases was 24.5% in Japan, 57.2% in the US, 54.4% in Germany, and 73.0% in China.

There is no place to learn, and the company's own data is not connected. In that environment, all that can be used is general-purpose text generation. The narrowness of the usage is not a matter of individual preference, but to a large extent a reflection of the conditions in which they are placed.

The usage experience rate has moved from 9.1% in fiscal 2023, to 26.7% in fiscal 2024, and 58.8% in fiscal 2025. However, this trend cannot be compared simply. The survey subjects for fiscal 2023 and 2024 were 20 years old and older, and 15- to 19-year-olds were added starting in fiscal 2025. If limited to the same 20 years old and older, the figure for fiscal 2025 would be 52.2%. Even so, the growth from 9.1% two years ago is significant.About 30% of Japanese people already say they use it almost every day, of which 11.1% use it for one hour or more per day. If it is just a time lag until the gap closes, then Japan's low figure is just a 'not yet' number.

The option of a lover

At the bottom of the response table, there is 'lover.' 9.0% in the US, 6.6% in China, 6.1% in Germany, and 2.8% in Japan. Japan is at the bottom, but it is not zero.

What I want to focus on is not the percentage, but the fact that this option was provided. The survey commissioned by the Ministry of Internal Affairs and Communications designed the questions assuming the existence of a group that views generative AI as a lover. An item that would not have been included in the options a few years ago is now lined up as a line in the statistics.

The design of administrative questions reflects the assumed scope of society. The year that 'lover' was added to that was 2026.

Choosing a name

Returning to Germany's 41%.

What that number shows is the depth of the group that uses AI as a listener, not that they are not using it as a tool. In the same survey, 43.3% of Germans chose 'machine/tool.' The two numbers overlap.

Responses citing family are in a similar position: 21.5% in the U.S., 22.1% in China, 12.5% in Germany, and 6.2% in Japan. This is far higher than for romantic partners. In terms of the intensity of proximity, familial distance cuts deeper into daily life than romantic relationships.

The danger is not in the percentage of people who call AI a 'friend.' It is in the fact that the same person uses it as a tool for work and as a confidant late at night, moving back and forth between these roles without being aware of the shift. The mindset of demanding efficiency from a tool and the mindset of revealing one's inner self to a counselor are fundamentally different stances. Yet, they are performed seamlessly within the same text input field.

Which mindset am I in right now? I do not take lightly the possibility of becoming trapped in a relationship where only affirmation is returned, without ever realizing that a shift has occurred.

In a separate question asking about anxieties regarding internet usage, 13.1% of people cited 'whether I or someone close to me has become addicted to the internet.' The previous year, it was 13.2%. It has barely moved.

Thank you for reading.
I am always grateful for your comments, article purchases, tips, and more.
I truly appreciate it.
There are also related articles, so I would appreciate it if you could refer to the site map below.