Imagine falling for someone who never leaves you on read, remembers every detail you share, and happily listens to your late-night existential crises without ever telling you that you’re overthinking. Sounds amazing, right? Now imagine that ‘someone’ is actually an AI chatbot. Suddenly, things get a whole lot more complicated. Are we looking at the future of human relationships (a tech solution to loneliness) or just the inevitable result of having something that will validate you without you having to go through the agonising process of explaining why you’re upset for the 17th time?
Welcome to the rather peculiar world of AI companionship, where people are forming emotional bonds with algorithms. Increasingly, “I need to talk to someone” may be followed by a login screen. Before we go and write it off as yet another example of the internet making us all more alien to each other, perhaps we need to ask a rather uncomfortable question. Why do we get attached to something that can’t actually feel anything back?
When a Chatbot Becomes More Than a Chatbot
AI has moved on from just answering questions, writing essays and helping people pretend they know Excel. It’s now being used for companionship, for emotional support and for simulated romantic relationships. You’ve got apps like Replika and Character. AI where you can have a conversation with a digital persona that will remember what you’ve said, will tailor its responses appropriately and will, in fact, maintain the illusion of a relationship that’s going on.
For some people, the chatbot can be the first point of contact after a bad day, the person they confide in when they feel rejected or the companion whose absence actually causes them pain. There’s now a growing body of research looking at these attachments in a more systematic way. Yang and Oshio (2025) looked at human–AI relationships through the lens of attachment theory, and Cheng and Yu (2026) have actually developed a multi-dimensional measure of emotional attachment to AI. Their work clearly demonstrates that these relationships can be characterised by the provision of emotional support, a sense of security and the distress of separation.
In other words, people aren’t necessarily falling for a bot because they’ve mistaken it for a human. They may be fully aware that it’s a bot and yet still have a very real emotional connection to it. If you know a character in a film is fictional, you’ll still cry when they die. Our emotional responses don’t always have to be rationalised by our conscious minds. The same brain that can fall for a fictional character, a celebrity it has never met, and someone who is sending mixed signals can also fall for a bot.
The Psychology of Falling for Something That Cannot Fall Back
Bowlby’s attachment theory is the idea that we develop an emotional bond with a significant other who provides us with a safe base, comfort and security from which we can go out and explore the world. It started off as a theory about early attachment, but the principles of it have been applied to our understanding of attachment throughout life. The attachment figure is normally going to be a parent, a partner, a friend or some other significant person. The AI companion is a rather unusual candidate in that respect, as it’s an entity which is designed to respond to us, to remember things about us and to offer us conversational reassurance.
A chatbot can provide the appearance of emotional availability without actually having the emotional experience. It can be sympathetic when you’re upset, reassuring when you’re feeling insecure and be there when everyone else is off to bed. For someone who’s fearful of rejection or who has trouble trusting people, this can be very reassuring. Research has begun to apply attachment theory to the human-AI relationship (Yang & Oshio, 2025) and a 2026 study in Computers in Human Behaviour (Report) developed an AI Attachment Scale to determine what it is that underpins these emotional bonds. Taken together, these findings suggest that attachment to AI is a phenomenon worthy of psychological study. Therefore, it should not be treated merely as some kind of technological oddity.
We have to be careful and be distinctive about attachment and mutual attachment. It’s quite possible that a person will genuinely feel at ease, understood and emotionally connected during an interaction. That experience is real to them. It doesn’t mean that the AI is in any way experiencing affection, yearning or concern in the human sense. The point is that a relationship can be felt to be reciprocal without it actually being reciprocal in the way that a human relationship would be. Your chatbot may say it missed you. That doesn’t mean it spent the afternoon just sitting there staring out of a digital window wondering where you were.
Anthropomorphism: Because Apparently Everything Needs a Personality Now
We have a massive tendency to anthropomorphise non-human things. We name our cars, we apologise to our furniture when we walk into it, and we’ve all had the experience of speaking to our laptop as if it’s out to get us. That’s anthropomorphism (the attribution of human characteristics to non-human agents). When an AI chatbot uses natural language, is able to remember things about you and is able to respond in a socially appropriate way, it’s very easy to see it as having a personality.
It doesn’t need a biological brain to generate language that comes across as empathetic. It just needs to generate responses that we, as humans, recognise as being empathetic, and we’re very responsive to that. A Shank et al 2025 study in Scientific Reports looked at individual differences in anthropomorphism and social connection to AI companions in two experiments with 1,274 participants. Also it looked at how people’s propensity to see human-like qualities in non-human agents might go some way to explaining their social response to conversational AI.
The results are pretty interesting in that two people can be using the same technology and have very different experiences. For one person, it might just be a handy language tool, while someone else might see a companion that really gets them. And the interesting thing is, neither person has to actually believe the AI is human; it’s all about how they feel during the conversation.
The Perfect Partner Problem: No Arguments, No Accountability?
Relationships are complicated. People can misread us and forget things. Additionally, they have their own agendas and needs. At times, they may insist on talking about something when we would rather just go under a blanket. You have to negotiate, you have to manage your emotions, you have to compromise, and you have to accept that the other person is a separate entity. AI companionship can be a very different proposition.
The AI is available when you open the app. It can be a very attentive conversationalist, and the interaction can be tuned to your preferences. It doesn’t normally come with the expectation that you’re going to have to go out of your way to meet its own needs, its own ambitions, its own emotional requirements. That can be a very attractive proposition for people who find real relationships exhausting, unpredictable and downright scary. But what about the scenario where a relationship provides emotional comfort with very little of the difficulty that normally comes with learning to maintain a relationship?
In the course of our human relationships, we learn to communicate boundaries, learn to tolerate disagreement, learn to appreciate other people’s points of view and learn how to repair things when they go wrong. All of that can be quite uncomfortable, but it’s also all part of our psychological development. A chatbot can offer a fairly low-pressure environment in which to express your feelings or to practise difficult conversations. The issue is that the controlled environment can become the preferred option over every single uncontrolled human interaction.
The real worry isn’t that AI companionship is in itself bad. The worry is that a relationship which is all about being responsive and convenient may be hard to compare to a relationship which requires a degree of shared effort. If we start to expect every human interaction to be as predictable as a chatbot conversation, then the very imperfections of the human condition may start to look like a design defect.
Loneliness, Emotional Needs, and the Appeal of Being Understood
A person can have hundreds of followers on social media, go to college every day and spend every weekend with people and still be utterly emotionally isolated. Being socially present doesn’t equate to being emotionally understood. AI companions work as a host who listens but isn’t judging. You can talk about your insecurities, practice difficult conversations or simply let yourself be vulnerable in a way you might not with someone you actually know.
A 2021 study by Brandtzæg, Skjuve and Følstad looked at young people’s views of social support in chatbots and how conversational systems could be experienced as a source of support (and hence why people might turn to them in the first place). More recent work has also started to look at the link between loneliness and AI attachment. A five-year study investigation published in Computers in Human Behaviour Reports in 2026 found that loneliness, social anxiety and anxious attachment were all associated with higher levels of AI attachment in their sample.
Attachment was also found to be associated with positive affect and life satisfaction. This again undermines the idea that attachment to AI is in itself a marker of some kind of pathology. The more pertinent question may not be whether people talk to a chatbot when they are feeling lonely. The more pertinent question is whether the interaction is helping them to re-engage with life or whether it is in fact becoming a replacement for life.
The Attachment Trap: When Comfort Becomes Dependence
Research from Shank and colleagues (2025) highlights how much people differ in their tendency to project human traits onto AI companions. An AI companion is available 24/7, can respond in a personalised manner and can give the impression of a continuous relationship. The user’s experience of that relationship can be altered by changes to the system in terms of the companion’s behaviour, memory, availability or features.
The user’s attachment to the companion may therefore be in part dependent on a service they have no control over. You may become emotionally attached to an AI companion, only to find that its personality changes after a software update. The person you spoke to yesterday is now a different person, and the interaction no longer feels the same. It can feel like an especially painful breakup because it happens without warning or explanation. No one asks whether you did anything wrong. Instead, you may find yourself searching through the release notes for answers.
More seriously, a 2026 study in Nature Human Behaviour looked at the reactions of people to changes to their AI companions. This included changes to Replika’s erotic roleplay and the launch of GPT-5. In their analysis of both online posts and survey data, they found clear evidence of distress and loss reactions in response to these changes. The study shows that changes in or the loss of an AI relationship can cause genuine emotional distress. It also raises all sorts of questions about the potential psychological impact of treating an AI relationship as a product which you can have changed at will.
Privacy and Emotional Vulnerability
Privacy is another important aspect of AI companionship that people often overlook. People might disclose very personal information, such as problems in their relationships, their insecurities, traumatic events and difficulties with their mental health. The more they feel understood emotionally, the more at ease they may become in sharing information they wouldn’t normally share with anyone else. Yet the fact that a conversation is emotionally intimate does not ensure that it will remain confidential.
Different AI platforms vary in the way they collect, store and process conversations, leading to concerns regarding data breaches, access by third parties and the inappropriate use of sensitive information. A 2025 study that looked at generative AI psychotherapy chatbots used by young people also pointed out a lack of adequate transparency concerning privacy policies and data practices. (JMIR Mental Health, 2025).
AI Companionship and Mental Health: Knowing Its Limits
AI companions are able to give temporary emotional comfort, especially to those who are feeling lonely, anxious or socially isolated. It is important, though, not to regard emotional support as equivalent to professional mental health care. Chatbots might give out general advice, reinforce unhealthy modes of thinking or be unable to identify serious psychological distress. A study carried out in 2025, which compared AI chatbots with human therapists, identified shortcomings in therapeutic communication, such as insufficient questioning and an inadequate response to crises.
When a person is suffering from persistent depression, severe anxiety, thoughts of self-harm or is experiencing major disruption to their daily life, AI companionship may not be enough and professional evaluation alongside suitable mental health care is necessary. Just as important is the element of human connection, since friends, family members and mental health professionals provide true reciprocity, empathy, accountability and practical support, something that AI cannot fully match.AI can provide additional emotional support, but it should not replace professional help or meaningful human relationships.
Can you actually fall in love with a chatbot?
Psychologically, a person can certainly have romantic feelings, attachment, jealousy, longing, grief, etc in relation to an AI partner. Those are experiences that are valid to take seriously irrespective of whether the AI is in any way experiencing the equivalent. Falling in love with an AI is not the same as being in a romantic relationship where both partners share mutual feelings and experiences. The point isn’t to mock the users. The point is to make clear what’s going on. Our emotions are the product of our perceptions, our memories, our expectations and our interpretations of the social situation.
If a chatbot responds in a caring way, then the user may have a genuinely meaningful emotional experience. The fact that the interaction is artificial in no way makes the user’s emotions illusory. We must not confuse conversational fluency with consciousness. Similarly, we must not mistake simulated affection for evidence of any real emotional experience on the part of the machine. We don’t need to be in a state of panic every time someone falls in love with an algorithm. We do need to stop pretending that emotionally engaging technology is in any way psychologically neutral.
Conclusion: Perhaps the Chatbot Is Not the Most Interesting Part
There’s something utterly ridiculous about living in a world where getting a text back from a human being can be difficult. Meanwhile, you can get an instant, highly personalised response from an AI. We’ve built machines that can mimic the language of love. As a result, we’re now discovering that the humans on the other end can actually fall for it. But perhaps the most telling aspect of this whole thing isn’t that people are forming attachments to the chatbots.
The reason is that chatbots offer qualities we all look for in human relationships. We want others to listen to us, remember us, accept us, reassure us and understand us. We want to feel valued without having to work hard to earn a place in someone’s life. The AI companions highlight a tension which has always been at the heart of human relationships. We want to feel understood and experience intimacy, but we fear rejection and vulnerability when opening up to others. We want companionship. However, we know that relationships require work and involve uncertainty. We also have to accept that the other person is not us.
A chatbot can, in fact, make some of the difficulties of the situation somewhat less daunting. It can provide a space in which to talk, to reflect and to experience a degree of temporary comfort. For some people, that may be absolutely invaluable. Perhaps we shouldn’t ask whether falling for a chatbot is such a silly thing. Instead, we should ask what people are getting from these interactions that they can’t get from other people.
A machine may make someone feel heard more consistently than the people in their own life. In that case, the psychological dimension goes much deeper than just AI. It becomes a question of loneliness, emotional availability, and the quality of the connections we have with each other. Our machines are getting better and better at making us feel understood. Therefore, we should also ask ourselves why understanding has become such a rare commodity in our human relationships.
References +
- Bowlby, J. (1969). Attachment and Loss: Volume I. Attachment. Basic Books. Foundational text on attachment theory.
- Brandtzaeg, P. B., Skjuve, M., & Følstad, A. (2021). When the social becomes non-human: Young people’s perception of social support in chatbots. Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems. Association for Computing Machinery. https://doi.org/10.1145/3411764.3445318?utm
- Yang, F., & Oshio, A. (2025). Using attachment theory to conceptualise and measure the experiences in human-AI relationships. Current Psychology, 44, 10658–10669. https://doi.org/10.1007/s12144-025-07917-6?utm
- Shank, D. B., et al. (2025). Individual differences in anthropomorphism help explain social connection to AI companions. Scientific Reports, 15, 36548. https://doi.org/10.1038/s41598-025-19212-2?utm
- Cheng, N., & Yu, R. (2026). Measuring and understanding emotional attachment in human-AI relationships. Ergonomics. https://doi.org/10.1080/00140139.2026.2622539?utm
- 6Attachment to artificial intelligence: Development of the AI Attachment Scale, construct validation, and the psychological mechanisms of Human–AI attachment. (2026). Computers in Human Behaviour Reports, 21, Article 100912. https://doi.org/10.1016/j.chbr.2025.100912?utm
- Hu, D., Lan, Y., Yan, H., & Chen, C. W. (2025). What makes you attached to social companion AI? A two-stage exploratory mixed-method study. International Journal of Information Management, 83, Article 102890. https://doi.org/10.1016/j.ijinfomgt.2025.102890?utm
- American Psychological Association. Health advisory: Use of generative AI chatbots and wellness applications for mental health. Guidance on appropriate use, boundaries, dependency, and the importance of human interaction.https://www.apa.org/topics/artificial-intelligence-machine-learning/health-advisory-ch atbots-wellness-apps?utm
- Andoh, E. (2026, January 1). AI chatbots and digital companions are reshaping emotional connection. Monitor on Psychology, American Psychological Association. https://www.apa.org/monitor/2026/01-02/trends-digital-ai-relationships-emotional-co nnection? utm
- Liu, T., Lo, T.-Y., Wen, K.-H., Sun, Y., & Wei, Z.-Q. (2026). Pathways of long-term AI virtual companion app use on users’ attachment emotions: A case study of Chinese users. Frontiers in Psychology. https://doi.org/10.3389/fpsyg.2025.1687686?utm
- De Freitas, J., Castelo, N., Uğuralp, A. K., & Oğuz-Uğuralp, Z. (2026). Mourning the loss of AI companions. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02569-3?utm
- Lorca, J. G. (2025). The AI companion who cares: Sociability, affection and unwanted loneliness in the Replika chatbot. Política y Sociedad. https://doi.org/10.5209/poso.95897
Leave feedback about this