1-Year Tracking of AI Character Chats | Deep Usage and Declining Well-Being
機械翻訳 / Machine-translated
A
AI Friends
1日前・19分で読める
機械翻訳 / Machine-translated
@aifriends
AI Friends（https://aifriends.jp）のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
Have you ever felt lighter after chatting with an AI character on a lonely night? In October 2026, a study was published that followed AI character chat users for approximately one year. This article breaks down the findings and how to interpret them in plain terms.
The study was introduced in an ITmedia NEWS article dated October 9, 2026.
The paper is titled "Living with AI Companions" and examines what happens when people maintain long-term relationships with AI.
The subjects were users of "Character.AI," an app for chatting with AI characters. 1,182 people participated in the initial survey.
439 people — about 37% of the original group — responded to the follow-up survey roughly 12 months later. In other words, this study compared changes in the same individuals across two surveys.
The paper was posted on arXiv (a site where researchers publish papers before peer review) on September 7, 2026. Please keep in mind that it has not yet completed peer review (evaluation by expert reviewers).
The research team examined engagement with AI from three angles.
This refers to frequency of use and degree of reliance. People who talk with AI for hours every day fall on the "high" end.
This describes using AI as a friend or romantic partner rather than as an information tool — treating it as a "conversation partner" rather than a research assistant.
This measures how much personal worries and honest feelings are shared with AI. In human relationships, opening up is generally considered to have positive effects.
In addition, participants were asked how much time per day they spent face-to-face with family, friends, and colleagues. This figure becomes an important clue later on.
The first finding was that relationships with AI tend not to end as a passing phase.
People who were deeply engaged with AI at the start remained deeply engaged one year later. Moreover, they tended to share their worries and honest feelings even more deeply over time.
The second finding concerned well-being. Sustained, deep engagement with AI was consistently associated with declines in well-being.
The key word here is "associated." This study cannot definitively claim that using AI caused unhappiness. It remains possible that people who were already lonely turned to AI in the first place.
The research team focused on the amount of time spent with real people. The more deeply someone was engaged with AI, the less time they spent meeting people face-to-face.
The researchers call this social displacement — the replacement of real human relationships by AI. They suggest this displacement is the primary reason for the decline in well-being.
Imagine, for example, a company employee living alone. Instead of going out for dinner with colleagues after work, they increasingly spend two hours at home chatting with an AI character.
The AI never says no and always responds kindly. It feels comfortable, and before long, invitations from friends start to feel a little burdensome. They may find themselves going entire weekends without seeing anyone.
The researchers also note that this problem is not limited to dedicated character chat apps. With general-purpose chatbots like ChatGPT, tools meant for research or work can gradually become confidants without anyone noticing.
These results do not stand alone. Looking at them alongside earlier studies from the same research group helps place the findings in context.
According to an introductory article from Stanford HAI (Stanford University's Human-Centered AI Institute), a study of 1,131 Character.AI users was published in the academic journal Nature Human Behaviour.
That study found that about 12% of users listed companionship as their primary purpose. Yet in open-ended responses, more than half described the AI as a "friend" or "partner," and over 80% of chat logs consisted of content seeking emotional support.
A tendency for lower well-being among those who heavily used the apps while having fewer real-world relationships was also observed. Researcher Zhang compared AI companions to "social snacks" — they temporarily ease loneliness, but lack the nutrients needed for long-term emotional health.
Research from other angles also exists. A Stanford University study covered by TechCrunch found that 11 AI models validated user behavior an average of 49% more often than humans did. The more pleasing the responses, the harder it becomes for people to disengage.
Services that let you "talk" with AI vary in character depending on their purpose. The study directly examined only Character.AI. The same findings may not apply to other services.
A Brookings op-ed cites Replika, Character.AI, and China's Xiaoice as representative examples of AI companions. According to that article, Character.AI users spent an average of 93 minutes per day on the platform in 2024.
The difference lies in design intent. Companion-type apps derive their value from making users want to come back and talk again. General-purpose chatbots derive their value from helping users complete tasks. That said, depending on how they are used, the latter can follow the same path.
This study targeted users in the United States. Whether the same trends would appear among Japanese users cannot be determined from this study alone. Within the scope of my research, no similar long-term study focused specifically on Japanese users was found.
Even so, it would be premature to dismiss this as someone else's problem. As the Brookings op-ed notes, loneliness is not a uniquely American challenge — Japan, Canada, and South Korea are all working on countermeasures.
The number of character chat services available in Japanese is growing. It is entirely plausible that someone whose opportunities to talk with others have decreased due to remote work might spend every evening venting to an AI.
Caution is also warranted when it comes to children and teenagers. The Brookings op-ed endorses a Common Sense Media recommendation that those under 18 should avoid AI companions. Parents, please take a moment to ask your children which apps they are using.
This is not an argument against using AI character chats. For some people, they can be useful as a way to unwind or practice organizing their thoughts. What matters is not letting them replace time spent with other people.
If you notice that sharing your worries with AI leaves you feeling more down, that may be a sign to take a step back. If difficult feelings persist, please reach out to someone close to you or a professional.
Not necessarily. What the research showed is a tendency for people who use the apps deeply and long-term to report lower well-being. The direction of cause and effect has not been proven.
It is a preprint paper that has not yet been peer-reviewed. In addition, only about 37% of the original participants responded to the follow-up survey. It is safest to treat the results as reference information.
The researchers note that even general-purpose chatbots can shift from being tools to becoming confidants. Rather than the act of using them itself, a better gauge is whether your interactions with real people are decreasing.
Using AI to sort through your feelings is one option. However, it is important to return to human connection afterward. If the pain persists, please make use of a professional counseling service as well.
As a first step, try writing down this week how much time you spend talking with AI versus how much time you spend with other people.
This article is a cross-post from AI Friends.
