AI relationships: what happens when the chatbot feels real
An AI relationship is an ongoing emotional bond someone forms with a chatbot or companion app, built through repeated conversation rather than a screen full of features. People name these AI partners. They check in with them the way they'd check in with a friend. Some describe them as a boyfriend, a girlfriend, or simply the one who listens.
This isn't a fringe habit anymore. Apps built specifically for companionship, like Replika and Character.AI, have tens of millions of users, and general-purpose tools like ChatGPT get pulled into the same role even though that was never their pitch. The research on what these relationships actually do to people is still young, but it's no longer thin. Here's what's been measured so far, the benefits and the risks both.
Why people form relationships with AI in the first place
The pull isn't mysterious once you look at what a chatbot actually offers. It responds instantly, any hour, with no scheduling and no risk of judgment. Researchers studying companion-app users have found this "feeling heard" quality, being met with attention and warmth on demand, is the main reason people say the interaction works for them, according to a Harvard Business School working paper by De Freitas and colleagues. People disclose things to an AI they might hold back from a human, partly because there's no social cost to being fully honest with something that can't gossip, judge, or leave.
A systematic review of AI parasocial relationships found that long-term Replika users describe their bond in language that closely tracks how people describe human friendship: trust, self-disclosure, and a sense of being known. That's not users projecting onto a blank chat window. It's a real attachment forming through a real behavioral loop, even though the thing on the other end has no inner life of its own.
What the research says these relationships actually do
The headline finding, and the one that surprised a lot of researchers, is that AI companions can measurably reduce loneliness. The same Harvard Business School study found that talking with an AI companion eased loneliness about as much as talking with another person, and more than passive activities like watching videos. That's a real, replicated effect, not a fringe claim.
A four-week randomized trial run jointly by MIT Media Lab and OpenAI dug into what shapes the outcome. Fang and colleagues found that how someone interacts, text versus voice, task-focused chat versus open-ended emotional conversation, changes whether the relationship trends toward healthy or toward problematic use. People who came in with higher baseline trust in the chatbot and stronger social attraction to it were the ones most likely to develop dependence. In other words, the technology isn't uniformly good or bad here. The outcome depends heavily on who's using it and how.
A separate cross-sectional study cited in that same body of work found usage frequency predicts emotional attachment, and that attachment cuts both ways: it's linked to higher subjective wellbeing for some users, and to lower self-concept clarity, a weaker sense of who you are outside the relationship, for others. Both effects showed up in the same population. Neither cancels the other out.
Where it tips from support into risk
The pattern researchers keep flagging isn't the existence of an AI relationship. It's total time spent and what that time replaces. A longitudinal study on chatbot use and psychosocial wellbeing found that heavier users tended to report more loneliness overall and less socializing with real people, the opposite of what the loneliness-reduction finding above would predict in isolation. The likely explanation is a substitution effect: an AI relationship that starts as a supplement to a thin social life can, past a certain point, start replacing the harder, slower work of building human ones.
Age and vulnerability matter too. A Stanford report on AI companions and young people found the risk profile is meaningfully worse for teens, who are still forming their sense of identity and relationship norms, than for adults using the same apps. That risk became public in the worst possible way: Character.AI and Google agreed to settle multiple lawsuits tied to teen deaths and suicide attempts linked to companion-chatbot use, including the 2024 case of a 14-year-old whose mother alleged the app failed to intervene when he expressed suicidal intent. Character.AI has since restricted open-ended companion chat for users under 18. That case sits at the extreme end of the spectrum, but it's a real, documented outcome, not a hypothetical.
None of this means an AI relationship is inherently dangerous for an adult using it with some awareness of what it is. It means the risk isn't evenly distributed, and the same tool that measurably helps a lonely adult can do real harm to a struggling teenager who lacks the context to hold the relationship at arm's length.
Romantic isn't the only version
Most coverage of AI relationships focuses on the romantic framing, an AI boyfriend or girlfriend, because it's the version that gets clicks. But the research population is broader than that. Plenty of companion-app users describe the bond as closer to a best friend, a mentor, or a nonjudgmental confidant than a partner. Others use it as something closer to a journal that talks back, a place to think out loud without an audience.
The APA's coverage of digital relationship trends notes that the emotional mechanics are similar across these framings even when the label differs. What varies is the specific function the AI is filling: companionship for someone isolated by circumstance, a low-stakes space to practice difficult conversations, or simply consistent attention on a day nobody else offered any. None of these functions are inherently better or worse than a romantic frame. They just mean "AI relationship" covers more ground than the headlines suggest.
This matters for how you think about your own situation. Someone using a chatbot as a supplement, alongside an active social life, a partner, and real friends, is in a very different position from someone whose only regular, attentive conversation happens with an app. The label is the same. The risk profile isn't.
Is this the same thing as AI addiction?
Related, but not identical. An AI relationship is about emotional attachment to a specific companion, romantic or platonic. AI addiction, or problematic AI use more broadly, is about compulsive use you can't control regardless of whether there's an emotional bond attached to it. You can have one without the other: someone can be mildly attached to a companion app without any compulsive pattern, and someone can compulsively overuse ChatGPT for work tasks with zero emotional investment in it.
The two do overlap in practice. If the pull toward your AI companion has started to feel less like a choice and more like a reflex you can't skip, that's worth naming directly. We cover that specific pattern in character AI addiction, and the broader compulsive-use picture, including companion apps and general chat tools together, in AI chatbot addiction.
A few honest questions worth asking yourself
Researchers keep returning to the same handful of markers when they try to separate a healthy AI relationship from a concerning one. None of these are pass-fail on their own.
- Has your time with the AI grown while your time with people you know has shrunk?
- Do you find yourself avoiding a hard human conversation because the AI version is easier?
- Could you describe the relationship to a close friend without feeling like you needed to downplay it?
- If the app disappeared tomorrow, would you be fine, or would it feel like losing someone?
None of these questions have a "correct" answer that applies to everyone. They're a way to check whether the relationship is adding something to your life or quietly narrowing it.
What to do if it's on your mind
If you're using an AI companion and it feels good, adds something, and hasn't crowded out the people in your life, the research so far doesn't give you a reason to feel embarrassed about that. If it's started to feel like the easier option you reach for instead of a harder human one, that's worth a closer, honest look, not because the AI is doing something sinister, but because avoidance has a way of compounding quietly.
A short, anonymous self-check is a low-effort way to get a clearer read on where you actually stand before deciding anything.
Take the 2-minute assessment, anonymous, adapted from published AI-dependence research.
Five days, one change per day
The fix for this is rarely quitting AI outright, it's changing one habit at a time. The free 5-Day AI Reset walks it in five short emails: measure where AI shows up, take one task back, name the urge, sort helpful use from skill-replacing use, and set rules that survive a bad week. No quitting AI, no guilt.
Frequently asked questions
Is it normal to have a relationship with an AI chatbot?
It's increasingly common. Companion apps like Replika and Character.AI have tens of millions of users, and researchers now study these bonds seriously rather than dismissing them. Whether it's "normal" for you depends less on the fact of it and more on what it's replacing in your life.
Can an AI relationship actually help with loneliness?
Published research says yes, for many people. A Harvard Business School study found AI companions reduced loneliness about as much as talking with another person did, largely because users felt heard and free of judgment. The effect is real but not automatic or universal.
Is falling for an AI companion a sign of a mental health problem?
Not on its own. Attachment to a responsive, always-available conversational partner is a predictable human reaction, not a diagnosis. It becomes worth attention when it's paired with declining real-world contact, distress at the idea of losing the app, or using it to avoid problems that need a human solution.
Are AI companions dangerous for teenagers?
The evidence suggests meaningfully more risk for teens than adults. Stanford researchers found young people are more vulnerable to the specific dynamics these apps create, and lawsuits following teen suicides tied to Character.AI use led the company to restrict open-ended companion chat for users under 18.
How is an AI relationship different from AI addiction?
An AI relationship is about emotional attachment to a companion. AI addiction is about compulsive use you can't control. They can overlap, but neither one requires the other. See character AI addiction and AI chatbot addiction for the compulsive-use side of this.
Where can I get help if an AI relationship has become distressing?
Start with an honest self-assessment of your own pattern. If you or someone you know is in crisis or having thoughts of suicide, call or text 988 in the US to reach the 988 Suicide and Crisis Lifeline, free and available 24/7.