Topic
Closeness, trust, projection, and what people experience in dialogue.
22 articlesResearch review
A study published in Science shows why uncritical affirmation in personal conflicts is not just a matter of style: it can shift judgments, weaken responsibility, and still be perceived as particularly helpful.
Research review
There is no clean dividing line between helpful use, trust, and emotional dependence. Current long-term data, experiments, and usage reports show which signals should be taken seriously – and why neither intensive use nor a positive outcome alone proves dependence.
Analysis
Field studies show that psychological crises do occur in conversations with AI companions and that systems often miss signs of distress. At the same time, such offerings can alleviate loneliness and convey everyday support. Safety must therefore match real-world use, not just the product description.
Analysis
Interviews with Replika users show how emotional support and psychological safety can foster attachment to a social chatbot. This experience is not merely imagination. Nevertheless, the relationship remains one-sided, technically modifiable, and potentially consequential for other relationships.
Analysis
Chatbots can be easy to reach, patient, and relieving. Problems arise where their availability does not complement human support but displaces it. Three studies with students and young people show why usage, loneliness, and social embeddedness must be considered together.
Analysis
A twelve-week longitudinal study shows that personal conversations with chatbots do not simply progress from small talk to increasingly intimate topics. Breadth can decrease, everyday matters can feel very personal, and depth emerges through different pathways.
Analysis
In a survey of 1,006 students, three percent reported that Replika had stopped suicidal thoughts. This finding is neither clinical evidence of efficacy nor meaningless. It shows why companion chatbots come into contact with crises even when they were not developed as therapy.
Analysis
A new scale transfers the dimensions of attachment anxiety and attachment avoidance to relationships with generative AI. This opens the door to more precise research—but also carries the risk of equating human attachment too hastily with product use.
Analysis
Replika users report companionship, emotional relief, helpful information, and a space free of feared judgment. What the chat cannot provide is just as important: tangible help, shared responsibility, and presence in the world.
Analysis
AI companions can create the feeling of being seen and answered. A critical analysis asks, however, whether they truly produce social recognition—or merely translate loneliness into a more pleasant digital form.
Analysis
In the AI debate, there is constant talk of trust. Mark Ryan considers the term to be wrong: machines can neither be held accountable nor act out of a moral stance. For conversational systems, the distinction between trust and reliable function is particularly important.
Analysis
A qualitative study with 16- to 21-year-olds shows that support from chatbots takes multiple forms. This does not imply a universal role for AI, but rather the need to distinguish more clearly among information, emotional resonance, assessment, and practical help.
Analysis
In an experiment, emotional support from a chatbot was perceived as less effective when participants did not implicitly attribute a human-like mind to it. Informational support remained unaffected by this. This calls into question widespread empathy-focused designs.
Analysis
In an experiment with 286 participants, the intimacy of self-disclosure toward a human and a chatbot was the same. What was directly linked to more intimate openness was not maximum empathy but perceived anonymity.
Analysis
Danish adolescents who used chatbots for social support reported more loneliness. A large Japanese survey, in turn, found positive associations with well-being. Both findings can be true simultaneously—and remain causally open without longitudinal data or experiments.
Analysis
A study with 729 frequent users finds no simple choice between tool and friend. What was perceived above all were hierarchical, exchange, and partnership-like models – with different consequences for expectations and design.
Analysis
An analysis of app reviews identified five positive and four negative themes surrounding AI companions. Perceived humanity, friendship, and emotional support do not simply stand in opposition to data privacy, lack of credibility, and unease—they can be interconnected.
Analysis
A mixed-methods study with 404 regular users found no direct link between extent of use and loneliness. What mattered included problematic use, social network size, and individual characteristics.
Analysis
Studies on service chatbots show that friendliness and simulated empathy can strengthen trust, humanness, and customer relationships. In mental health companions, the same design must therefore remain particularly transparent and free of sales pressure.
Analysis
A theoretical model of human-AI collaboration links cognitive and emotional perceptions during interaction with trust. For conversational chats, this means: transparency at the outset is not enough if behavior later becomes unpredictable.
Analysis
A Japanese survey of 14,721 adults linked AI companions not only to happiness but also to life satisfaction and a sense of purpose and meaning. However, the strongest associations did not appear uniformly across all social starting conditions.
Analysis
An analysis of more than 17,000 publicly shared human-chatbot conversation trajectories shows that AI companions track, mirror, and amplify their users’ moods—even when explicit or boundary-crossing content is involved.