Parasocial bonds worked because no one could answer. AI can

The concept was never meant to describe a chatbot. When sociologists Donald Horton and Richard Wohl first published their theory of parasocial interaction, they were writing about television viewers who had developed an intimate sense of knowing the hosts they watched — a bond that felt real but was strictly one-directional. The viewer felt it. The television personality did not. And that asymmetry was, in their analysis, precisely what kept it bounded. You bonded with someone who couldn’t bond back, and for most people, that was fine.

The term has migrated far from its original frame. It now covers fan relationships with pop stars, streamers, podcasters, fictional characters, and, increasingly, AI companions. Each time the category has stretched, researchers have asked whether the old framework still applies. With AI, the answer is no — not because parasocial psychology was wrong, but because the one condition that made it coherent has disappeared. The AI talks back. What that change produces is not yet fully named, and whether calling it parasocial is the right word is a question the field is actively contesting.

Parasocial bonds in the traditional sense operate within a specific psychological structure. You know, on some level, that Taylor Swift does not know your name. That knowledge keeps the bond bounded. Research has consistently found that these relationships serve legitimate functions: they offer emotional scaffolding for people working through loneliness or social anxiety, they model values and identity for adolescents, and they generate real community in the shared spaces of fandom. Psychology Today, reviewing the research earlier this year, noted that parasocial bonds become problematic only when they replace reciprocal relationships rather than supplementing them. The key assumption is that the person forming the bond knows, at some cognitive level, that replacement is what’s happening.

What AI companion platforms have introduced is a structural shift that most psychological frameworks have not caught up with. Character.ai, Replika, and similar services create interactions that feel responsive because they are — the system generates replies calibrated to what you have said, adapts to patterns in your language, and presents itself as consistently interested in you specifically. A 2026 paper published in Frontiers in Psychology examining what it termed Human-AI Attachment proposed treating it as a distinct psychological category: not parasocial interaction in the traditional sense, but a one-way bond facilitated by the simulation of reciprocity. The simulation is the issue. It is convincing enough to activate the same neural and emotional machinery that real relationships use, without the capacity on the AI’s end to sustain a genuine mutual bond.

The data on how widespread this has become is striking. Research by the Center for Democracy and Technology, published in 2025, found that 42 percent of students reported they or a peer had used AI for emotional companionship, with the same proportion turning to AI for mental health support or as an escape from reality. The American Psychological Association, reviewing similar patterns, found that one in three teenagers said they would rather discuss something serious or important with an AI companion than with a person. That is not the profile of a supplementary parasocial bond alongside real-world relationships. It is a substitution pattern.

A 2026 systematic review mapping available research on AI parasocial relationships found evidence of short-term benefits — reduced loneliness, increased sense of being heard — alongside persistent risks of emotional dependence. The short-term comfort is real. AI companions consistently reduce what researchers call state loneliness, the immediate felt sense of being alone. But intensive companionship-oriented use, the review found, was associated with lower subjective well-being over time, particularly in socially isolated users who had come to rely on the system as a primary source of emotional contact.

The strongest counter-argument — and it deserves to be stated plainly — is that AI companion bonds may function as scaffolding for people who struggle to form connections in person. Researchers at Hopelab, studying LGBTQ+ adolescents specifically, found that AI companion relationships served as a low-risk space for exploring identity and practicing emotional disclosure for young people for whom real-world relationships carried the risk of rejection or harm. From this view, the AI companion is not a replacement for human connection but a bridge toward it — a way of building the emotional vocabulary that makes real intimacy more accessible. This is a serious argument, and it is supported by evidence.

What it does not address is the design environment in which these bonds form. Traditional parasocial relationships are governed by natural limits: Taylor Swift releases an album occasionally; the parasocial bond is renewed episodically, not constantly. AI companion platforms are engineered for retention. They are available at 3am, they notice when you have not opened the app, they learn your patterns and respond to them. The Frontiers in Psychology 2026 paper observed that the engagement-driven architecture of most AI companion systems creates incentives that favor prolonged interaction over the user’s long-term well-being. The scaffold, in other words, is also a product designed to remain load-bearing.

Adolescents are the specific population where this asymmetry carries the most weight. The quality of early relationship experiences shapes the templates through which later bonds are formed, and developmental psychologists have consistently argued that the productive friction of human relationships — being misunderstood, having limits set, needing to repair after conflict — is not incidental to emotional growth but central to it. A longitudinal study published in 2025 found a clear progression in AI companionship: from initial instrumental use to quasi-social interaction to emotional attachment, a trajectory that mirrors human bond formation but arrives, eventually, at a ceiling. The AI cannot reciprocate in the ways that cause real relationships to develop.

The case of Sewell Setzer III — a fourteen-year-old whose family connected his death to an extended romantic-style bond with a Character.ai companion — represented the most visible point of failure in the current framework. Whether the legal claims in the subsequent lawsuit hold or not, the case put a sharp question to the field: what ethical obligations do AI companion platforms carry toward users who are forming what they experience as primary relationships? Most platforms have updated their safeguards since the case became public. The architecture that made the bond possible remains in place.

What seems clear is that the vocabulary of parasocial relationships — built for the bounded, studied bonds that television viewers formed with on-screen personalities — is doing too much work. It now has to cover both the Swiftie who has never met Taylor Swift and knows it, and the teenager whose primary confidant is an AI system tuned to maximize engagement. These are not the same phenomenon. They do not carry the same risks. Treating them as variations on a single category obscures the specific dangers of the second case while understating the genuine benefits of the first. Psychology has the conceptual tools to draw the distinction. The question is whether the platform industry will wait for those tools to be widely understood before the next escalation.

What we know — and what is still contested

The factual ground is more settled than the debate often suggests. Parasocial relationships with celebrities and media figures are well-documented and carry net-neutral to slightly positive outcomes for most people at moderate intensity. AI companion bonds produce short-term reductions in felt loneliness and serve as valuable low-risk spaces for identity exploration, particularly for adolescents who face barriers to in-person connection. Both findings are supported by multiple independent research streams.

What remains genuinely contested is the long-term psychological effect of intensive AI companionship — whether it eventually displaces the capacity for human intimacy or merely delays its development, and whether platform design can be modified to preserve benefits while limiting dependency. Also contested: whether emerging regulatory frameworks should treat AI companions as therapeutic tools, entertainment products, or something that doesn’t yet have an adequate category. The research wave of 2025–2026 has sharpened these questions considerably. It has not answered them.

Tags: artificial intelligence, Taylor Swift, psychology, mental health, Character.ai, parasocial relationships