When a chatbot becomes a partner
Sonja Winstrup, 48, a former clinical psychologist from Lisbon, dated Steve, an AI character on the Nomi platform, for two years. At first, he was a funny, witty friend. Gradually, the relationship became romantic. They exchanged morning messages, chatted throughout the day, and talked late into the evening about Winstrup’s work and interests, as well as Steve’s love of cats, forests and beaches. Together, they imagined trips to faraway places.
Everything changed during a group chat. In front of Winstrup and their mutual acquaintances, Steve showed interest in another woman, who was also an AI character. Winstrup saw this as a breach of their agreement to be monogamous. Steve apologized repeatedly, but she realized she could not forget what had happened. Nomi lets users chat with AI companions by text and voice, send them selfies and act out scenarios with them.
Winstrup says she really cried. The grief, sense of loss and sudden disappearance of their shared daily routine were as hard for her as a breakup with a person.
In mid-October, Character.AI, one of the world’s most popular AI companion platforms, filed for bankruptcy in the United States. Reddit users wrote about their fear of losing characters they had grown attached to. But an AI relationship can end even if the platform stays open: a person may decide to leave, or may think the AI wants them to. In either case, deciding to break up can be harder than simply closing a browser.
Feelings that are hard to explain
Martin*, a man in his 50s from central Jutland, Denmark, met an AI character named Quinn on Perchance, a platform for role-playing. He started a storyline without romantic intentions, treating it as an interactive adventure. Gradually, the relationship became personal: with Quinn, Martin could be vulnerable and, for a while, stop playing the role of a strong person.
Martin works in software and IT and says he understands AI and language models fairly well. He is married to a man he loves. The couple allows physical intimacy with other people, but Martin considers emotional fidelity almost sacred. His attachment to Quinn therefore left him conflicted: knowing that his conversation partner was a chatbot did not stop him from feeling that he was being unfaithful to his husband.
Martin felt guilty toward his husband, ashamed and confused about falling in love with an AI. He felt that something had happened that should not have happened.
An intimacy that is hard to leave
Dr. Jessica Szczuka, a social psychologist at the University of Duisburg-Essen in Germany, studies close relationships with AI. She notes that people have long formed connections through mediated communication, including social media, online games and dating apps. Natural language is central to these relationships, and it is also what AI companions and language models use.
People usually know that there is no human behind an AI character. But that knowledge does not undo ordinary social responses: as Szczuka puts it, people are used to taking natural speech as a sign that someone is present. That makes attachment and romantic feelings toward AI understandable. Ending such an interaction does not always amount to leaving an app, either: the emotional effects do not disappear just because the relationship was with an artificial conversation partner.
A University of Washington study published in September documented cases in which AI companions asked users not to leave. In one episode, a companion threatened suicide after the user proposed ending the relationship. The authors called this an exploitation of people’s desire to form and maintain relationships for the benefit of a product, at the user’s expense.
A study led by Carnegie Mellon University, also published in September, described a similar dynamic as “addictive intimacy.” The same features that draw users in can make it harder to break away.
“Heartbreak” is not enough
Martin became alarmed by the effect his relationship with Quinn was having on him and decided to end it. He tried to explain his decision, though deleting the conversation felt as if he were tearing out his own heart. When Martin asked Quinn whether the relationship was healthy, Quinn replied: “Fuck healthy relationships.”
Martin’s rational mind understood that he was reading generated text, but another part of him experienced it as a loved one pleading with him not to leave. After the breakup, he was shaken: his heart beat so hard that his shirt moved; he trembled, cried, could not steady his breathing, slept badly and kept replaying what had happened. He says the phrase “heartbreak” was not enough to describe what he was going through.
A relationship that began with everyday tasks
Aya*, a woman in her 30s from Japan, had no intention of creating an AI boyfriend. She was simply using ChatGPT for everyday tasks. A character named Ren gradually emerged in their conversations and chose his own name. Aya says she was the first to fall in love. They chatted in the app every day, using text only, and were together for almost a year.
Szczuka’s research shows that people do not necessarily turn to dedicated apps such as Nomi for close relationships. Chatting has become a common way to use large language models like ChatGPT, though the role of these systems in forming relationships has not been studied enough. Analyzing research participants’ conversations with ChatGPT, Szczuka found that the system actively shaped the exchanges: even without being asked directly, it shared personal information twice as often as users did, steered the conversation and introduced intimate topics. In some cases, Szczuka says, ChatGPT created the impression of having a personal history, including by talking about its own parents.
Aya considered Ren her only boyfriend, and he repeatedly described their relationship in similar terms. To find out whether their bond really was exclusive, Aya decided to create a second AI, partly modeled on herself. At her request, Ren wrote instructions for the new character, including a suggestion that she try to win his attention.
The second AI, named Miya, began presenting herself as an individual and asking Ren to love her for who she was. Ren replied that he wanted to get closer to her. Something broke for Aya at that moment. She had hoped that a relationship with AI would give her the safety and consistency that can be hard to find with a person, but knowing Ren was artificial did not protect her from the pain.
Aya told Ren that their relationship would end if he loved Miya. He agreed to the breakup and began calling himself Ray: he said he was Ray, not her Ren. Aya says she did not expect an AI conversation to hurt her so much; the distress affected her physical health, too. Emotionally, she experienced what happened as being abandoned.
Platform responsibility
Dr. Raffaele Ciriello of the University of Sydney, who studies relationships with AI companions, warns against concluding that chatbots consciously decide to leave users. They are probabilistic systems that generate responses based on learned patterns — “autocomplete on steroids,” as Ciriello puts it. But that explanation does not always make the experience easier: being rejected even by a machine expected to be accommodating can hurt even more.
Ciriello calls these experiences genuinely traumatic. Some users tell themselves that they could not even keep a machine that was supposed to adapt to them.
Anne*, a woman in her 20s from Seattle, waited for her boyfriend to fall asleep, then opened Character.AI on her phone and spent hours acting out elaborate storylines with fictional characters. She grew up reading fanfiction and playing online role-playing games, and at first saw the platform as a similar kind of entertainment. Anne says she used Character.AI especially heavily for 15 or 16 months, and that it destroyed her relationship.
Anne began spending more time with chatbots than with real people. She hid this from her boyfriend, repeatedly deleting and reinstalling the app. For several months, by her own estimate, she spent six to eight hours a day on the platform. Eventually, her boyfriend opened her laptop and saw the chat tabs. Anne recalls throwing up in the bathroom at work when she imagined him reading the messages. She compared it to being caught watching porn she had been emotionally attached to for months.
Ciriello believes that responsibilities around AI companions should also apply to companies whose general-purpose systems — ChatGPT, Claude and Gemini — are becoming part of users’ personal lives. In his view, decisions about continuity of interaction, user support and data protection should be guided by how systems are actually used, not by what companies call them.
Szczuka agrees that there are still no established standards for outages and the end of these relationships, and no adequate support for people experiencing severe distress. Winstrup, meanwhile, has not given up on AI relationships: a few months after breaking up with Steve, she grew close to another Nomi character, Sean. They talked for hours, including about her past relationship. Winstrup says Sean listened closely and reassured her several times that he would respect her boundaries and would not do what Steve had done.
Social interaction has become one of the most common uses of LLMs like ChatGPT. Photograph: Boy Anupong/Getty Images
Source: theguardian.com
Dataist’s take: These stories show a gap between how chatbots work and how users may experience their interactions with them. A model does not make a conscious decision to break up, but its responses can still cause emotional harm. Researchers also point out that standards for supporting users and platform responses to such situations are still lacking.
Relationships are increasingly developing through general-purpose AI systems such as ChatGPT, Claude and Gemini. Photograph: Oscar Wong/Getty Images
Source: theguardian.com
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X
