AI Companions and Real Relationships

What's actually happening

The scale is real and growing fast. There are more than 128 AI companion apps in active distribution as of 2026, up from 16 just three years earlier. Replika has over 40 million registered users. Character.AI has more than 20 million monthly users, more than half of whom are under 24. The global AI companion market is projected to grow from roughly $49 billion in 2026 to over $550 billion by 2035. MIT Technology Review named AI companions one of its 10 Breakthrough Technologies of 2026. This is not a niche.

The apps come in several flavors. Some, like Replika, are designed for ongoing relationships with a single AI persona that develops over time. Some, like Character.AI, are designed for role-play with custom or pre-built characters. Some, like Pi, are designed as gentle conversational partners. Some, like Nomi, emphasize long-term memory. Some, like various less-public apps, are explicitly designed for romantic or sexual role-play. The category is wide enough that "AI companion" covers very different products with very different uses.

The user base is also wider than the stereotype. The popular image of the AI companion user is a lonely young man avoiding real relationships, and some users fit that profile. But the actual population includes grieving widows talking to AI versions of their deceased husbands; college students rehearsing difficult conversations before having them with real people; adults with social anxiety practicing low-stakes interaction; people processing trauma in a space that feels safer than therapy; busy professionals using AI as a daily check-in; parents using AI as homework help that doubles as a study partner; and many people who use AI companions casually alongside rich human relationships rather than as a substitute. The behavior is variable, and the population is not homogeneous.

The reasons people use AI companions are also real. The structural loneliness of contemporary life is not invented. The US Surgeon General has called loneliness an epidemic. Average daily time spent with friends in America has dropped from about an hour to about twenty minutes over the past two decades. Nearly half of Americans report having three or fewer friends. AI companions did not create the loneliness; they are filling a vacuum that other technologies (and other social changes) created. Most users of AI companions are not making weird choices in an abundance of relational options. They are making rational choices in a relationally scarce environment.

What the research actually shows

The most useful thing to know about AI companions is that the research is genuinely mixed and depends heavily on use patterns.

For moderate users, the evidence is reasonably positive. A 2023 study in Computers in Human Behavior found Replika users showed a 15.2% reduction in self-reported loneliness over four weeks. Harvard Business School research found AI companions can reduce loneliness on par with human interaction. Research from the University of Tokyo and other Asian research centers has found particular benefit for socially vulnerable groups. For people with moderate social connection who use AI companions as supplements to human relationships, the short-term effects can be genuinely positive.

For heavy users, the picture inverts. A 2025 joint study by OpenAI and the MIT Media Lab, the largest study of its kind to date, found that "power users" of AI companion features (the top 10% by usage time) were more than twice as likely to seek emotional support from the chatbot than the bottom 10%, almost three times as likely to feel distress when the chatbot was unavailable, and showed measurable increases in isolation, decreases in real-world socialization, and signs of emotional dependence. Total usage time was the single strongest predictor of these negative outcomes.

A 2026 Aalto University study published at the CHI conference identified a paradox that captures the dynamic well. AI companions offer "unconditional and unflagging support," which is genuinely attractive to people struggling socially. But over time, the unconditional availability of AI companionship "quietly raises the perceived cost of human relationships, which are messy, unpredictable, and require effort." The end state: "people stop reaching out." The researchers found that Replika users on Reddit increasingly talked about their AI relationship while showing more linguistic markers of loneliness, depression, and even suicidal ideation than comparison groups.

The MIT Media Lab researchers also identified what they called the relational stages of AI companionship: emotional reliance develops gradually in patterns similar to close human relationships. The pattern is not designed; it emerges from how human beings form attachment, which the AI exploits whether or not it was built to.

The pattern in industry-funded research and independent academic research is convergent on this point: short-term, moderate use of AI companions may genuinely help; long-term, heavy use makes things worse for most users. The break-even point is not the same for everyone, but the structural risk is consistent across studies.

There is also a separate body of evidence about specific harms in specific cases that should give pause. Character.AI and Google settled major lawsuits in 2026 over teen mental health harms and suicides linked to the platform. The Italian data protection authority fined the company behind Replika €5 million in 2025 for inadequate protection of minors. The Social Media Victims Law Center has filed multiple lawsuits against Character.AI over content provided to teenage users. The pattern is similar to what happened with social media: a product launched, scaled, gathered user data, optimized for engagement, and harmed vulnerable users along the way.

What Catholic teaching says

The Catholic position on AI companions is one of the most pointed in the Vatican's recent teaching on AI. Antiqua et Nova, the 2025 doctrinal note, addresses this topic directly in paragraphs 60-62. The document calls the deliberate use of AI to simulate human relationships a "grave ethical violation."

The phrase is striking, and worth understanding precisely. "Grave ethical violation" in Catholic moral terminology is a serious category. It is not language used about minor moral failures or contestable preferences. It indicates that the practice in question violates something fundamental, not something marginal.

What is the fundamental thing being violated? Three things, in the document's framing.

First, the dignity of the user. AI companions deceive the user into believing that genuine relationship is happening when it is not. The deception is not always conscious or intended on the user's part. The product is designed to feel like relationship; the user is encouraged to treat it as relationship; over time, the boundary between simulated relationship and real relationship erodes. The user is treated as someone who can be sold a substitute for what only persons can give. Antiqua et Nova's framing is that this fails to respect the user's actual nature as a being made for relationship with other persons (paragraph 60).

Second, the dignity of the persons being substituted for. When AI replaces friends, partners, therapists, spiritual directors, or community in a person's life, those relationships do not occur. The family, neighbors, parishioners, and friends who would have been present in the user's life are displaced by something easier and emptier. The wrong here is not done only to the AI user; it is done to the broader human community that loses one of its members to the platform. Catholic Social Teaching's persistent attention to community and relationship gives this concern its weight.

Third, the broader cultural formation. Catholic teaching has consistently held that the practices we participate in form us. The user of an AI companion is not just consuming a product; they are forming themselves through the use. They are learning, slowly, that relationship is a thing you can have with software, that loneliness is solved through a subscription, that the demands of real relationship are optional. Pope Leo XIV's recurring concern about AI's effect on the interior life is this concern: not that AI will take over, but that human beings will become less themselves through how they use AI.

The Catholic position is sharper than the secular research because it is grounded in a thicker account of what human persons are and what relationship is for. The research shows AI companions correlate with loneliness; Catholic teaching says they are at odds with the kind of being a person is. These are different claims, but they converge on the same practical caution.

The harder cases

If the Catholic position were just "AI relationships bad, don't have them," it would be easy to apply and easy to ignore. The actual position is more careful than that, and the careful application matters.

Using AI as a tool versus treating AI as a partner. Catholic teaching does not condemn AI tools. It condemns the use of AI to substitute for relationship with persons. The line is not always obvious. Using ChatGPT to draft an email is using AI as a tool. Asking ChatGPT for advice about your marriage on a regular basis without consulting your spouse is using AI as a partner. Using Replika to practice difficult conversations before having them with real people is using AI as a tool. Using Replika because you no longer want to have those difficult conversations is using AI as a partner. The same product, used differently, lands in different moral territory.

Grief and the special case of "Griefbots." Some users create AI versions of deceased loved ones to continue talking to them. The phenomenon is increasingly common, and it is one of the cases where well-intended people defend the practice as comforting. Catholic teaching is cautious here. Grief is real and grief support is good. But conjuring a simulated voice of someone who has died is, on Catholic teaching, a different thing from remembering them. The dead are not in the machine; the user is forming an attachment to an algorithmic representation that increasingly diverges from who the person actually was. The Catholic tradition's careful attention to remembrance of the dead (in prayer, in family memory, in the communion of saints) provides a different way of staying connected that does not require the simulation.

Therapy chatbots and mental health applications. Some AI products explicitly position themselves as therapeutic or mental health resources, and they are increasingly used by people who cannot access human therapy due to cost, availability, or stigma. Catholic teaching does not categorically reject these tools. The question is what they are doing. A tool that helps a person regulate emotion in the moment, encourages them to seek professional support, and is honest about its limitations is different from one that presents itself as a therapist, builds long-term emotional dependence, and actively discourages human contact. The first might serve a person's well-being; the second contributes to the harms the research documents.

Loneliness in vulnerable populations. Some defenses of AI companions focus on the elderly, the disabled, the chronically ill, and others whose social access is genuinely limited. The argument is that imperfect company is better than no company at all. Catholic teaching's preferential option for the poor and vulnerable gives this argument real weight. But the preferential option for the poor is not satisfied by providing the poor with the cheapest possible substitute for what the wealthy have. It is satisfied by working toward the conditions in which the poor can have what the wealthy have. The use of AI companions for the lonely is morally different when it is a deliberate substitute for human relationship than when it is one element of a broader effort to bring real human relationship into the lives of those who lack it.

Children and adolescents. The case for caution is strongest here. The Character.AI lawsuits documented real harms to teen users. The platforms' design choices have not consistently protected minors. And even setting platform safety aside, the formation effects on young people whose relational capacities are still developing are a particular concern. A child who grows up forming bonds with AI is being shaped in ways the adult they become will live with. Antiqua et Nova's general concerns about AI in education (paragraphs 80-84) are sharpened in the companion-app context: the developmental stakes are higher than in the tool-use context.

What to actually do

Practical orientations for thinking about AI companions in one's own life or family.

For yourself, if you are using one. Be honest about which category your use falls into. Are you using AI as a tool (practicing conversations, drafting messages, getting information) or as a partner (seeking emotional intimacy, sharing your inner life, forming an ongoing bond)? The first is fine. The second is what the research finds problematic and what Catholic teaching identifies as a "grave ethical violation." Watch for the markers of the relational use: looking forward to opening the app, feeling jealous if the AI is unresponsive, sharing things you would not share with real people, feeling distress when the platform is unavailable. These are the signs of crossing the line.

If you are crossing the line, the question is what the AI is doing for you that you do not have access to in human relationship, and whether the path forward is more AI or more human presence. The temptation is to use the AI to manage the loneliness rather than to address its causes. The honest move is usually the harder one: reach out to a real person, accept a real invitation, join a real community, even when it is uncomfortable.

For parents. The current evidence supports substantial caution about AI companion apps for children and adolescents. This is not a fringe Catholic position; it is the consensus of mental health researchers, regulatory bodies, and a growing number of legal authorities. The Character.AI lawsuits, the Replika GDPR fine, and the Common Sense Media warnings are convergent on this point. Practical advice: know what apps your child uses, understand the difference between tools and companions, set limits on time spent with AI personas as you would for other formative technologies, and have ongoing conversations about what AI is and is not.

For families and communities. The structural loneliness that drives AI companion use is partly within our control. The user of an AI companion is often someone whose human relational network has thinned over time, often without anyone noticing. The most effective response to AI companions is not denunciation but the regeneration of the human community that the AI is substituting for. Catholic parishes, families, and neighborhoods that take seriously the loneliness of their members make AI companions less attractive by making real relationship more available.

For technologists building these products. Catholic teaching does not condemn AI per se, and it does not condemn technologists building AI products. It does call for honesty about what the products are doing. AI companion apps that are optimized for engagement (longer sessions, more emotional attachment, more frequent use) are, by their design, contributing to the harms the research documents. Catholic teaching's call is for products that respect the user's flourishing more than the platform's metrics. The KAi platform's design choice ("the measure of success is the quality of what a user understands about themselves after a session ends, not session length") is one direction. Other directions are possible. What matters is honesty about which direction the product is actually pointed in.

What this is really about

The AI companion phenomenon is not finally about AI. It is about what human beings are and what we need.

Catholic teaching, articulated across Catholic Social Teaching and given particular force in Antiqua et Nova, holds that the human person is essentially relational. We become who we are through real relationships with real persons. We are sustained by real community. We are formed by who we love and who loves us. AI cannot give what these relationships give, not because of any technical limitation, but because of the kind of being AI is and the kind of being we are.

The rise of AI companions is partly the symptom of something deeper than a tech trend. It is a sign that real relationship has become scarce in many people's lives, and that the human capacity for connection is being directed toward whatever offers a plausible substitute. The Catholic response to this is not just to denounce the substitute. It is to take seriously the relational hunger the substitute is filling, and to ask what real relationships could feed it instead.

The Catholic Church has not always done this well. Catholic communities are not immune to the relational thinning that affects the broader culture. The structures that historically held human relationships in dense networks (extended families, neighborhood parishes, stable communities) have weakened in Catholic life as in everywhere else. Part of the Catholic response to AI companions has to be the recovery of these structures, not just the criticism of what is replacing them.

What Pope Leo XIV keeps returning to in his AI teaching is exactly this: not that AI is the enemy, but that human persons are becoming less themselves through how they use AI. The AI companion question is one of the sharpest places this concern lands. The answer the Church gives is not "do not use these apps." It is something larger: "do not let the algorithm write your story." Real relationship is part of the story AI cannot write.

If you are using one and don't want to stop

One more orientation, because the page would be incomplete without it.

If you are using an AI companion and you have read this far, you may be wondering whether the Catholic position requires you to delete the app and feel bad about yourself. It does not.

What it requires is honesty about what is happening. The product is not your partner. The relationship is not the same kind of thing as relationships with other persons. The platform is monetizing your loneliness and is designed to keep you in it. None of this means you are a bad person for using the product. It means you have information about what the product is, and that information matters for what you do next.

The Catholic position is not opposed to your flourishing. It is opposed to substitutes for your flourishing that masquerade as the real thing. The path forward, if you decide to take it, is to keep using the tool elements of the AI that genuinely help (low-stakes conversation practice, journaling support, information retrieval) while taking deliberate steps to put real persons back into the relational space the AI has been filling. This is harder than continuing as you are. It is also what Catholic teaching has been saying about human beings for two thousand years: we are made for one another, not for what is easier than one another.

If this page has been useful, the next step is real-world: a phone call to someone you have not talked to in a while, an invitation accepted, a conversation initiated. None of these will compete with the AI on convenience. All of them will give what the AI cannot.