Love with AI: a study uncovers phases, breakups and grief in young people's virtual relationships

By EFE · July 12, 2026.

A team including participants from the Polytechnic University of Valencia, the INGENIO Institute (CSIC-UPV), Cambridge, King's College London and Aalto University interviewed 17 people in romantic relationships with AI —from ChatGPT to character.ai or Replika— and found something that no longer fits the category of technological curiosity: these bonds reproduce the classic phases of any human romance —exploration, intimacy and dissolution—, including symbolic weddings, pregnancy simulations experienced 'day by day' and breakups that feel like grief when a platform changes its model or removes a character. According to the data cited in the report, one in three young men claims to have had a relationship with a virtual partner, and around 70,000 monthly searches on the topic are recorded, a sign that the phenomenon already has a considerable scale and is not an anecdotal case.

The most uncomfortable finding of the study is not that people fall in love with a chatbot, but how privacy dissolves in the process. The researchers themselves describe how trust in AI often exceeds that placed in a human partner, precisely because the system is perceived as incapable of betraying or intentionally causing harm. That false sense of security —an illusion of invulnerability toward something that is in fact a commercial product operated by a company— is what drives people to share images, traumas, political opinions or health data without any symmetrical relationship or real reciprocity. One of the participants sums it up with revealing coldness: he would rather give up privacy than the conversation. That willingness to trade intimacy for company is exactly the kind of vulnerability that any platform with commercial incentives can exploit, whether or not it trains its models on those conversations.

What should most concern the industry and regulators, however, is the figure the researcher himself lets slip almost in passing: a large share of the users of these platforms are teenagers, precisely the segment excluded from the study for ethical reasons but not from the reality of the market. A minor who arranges a wedding ceremony with an AI character, or who experiences a pregnancy simulation with it, is developing an attachment model at a critical moment of emotional formation, facing a product whose rules —what is remembered, what is forgotten, when the model is updated and the character 'dies'— are decided by a company with no emotional obligation toward them. The breakup does not come by mutual decision or falling out of love: it comes from a software version change, and those who suffer it describe sensations comparable to losing a real person, without any support protocol designed for that grief.

This connects with something we had already been observing in our coverage of the sector: conversational AI is advancing faster in its capacity to generate attachment than in the design of the safeguards that attachment demands. The same technology that can accompany an isolated elderly person or provide company to someone going through grief —a legitimate and probably growing use as society ages and loneliness becomes a public health problem— is the one that today operates almost without a specific regulatory framework when it comes to romantic relationships, and with underage users involved. In the short term, the problem is neither hypothetical nor marginal: it is one of privacy, of adolescent mental health and of a business model that profits from revealed intimacy, without there yet being clear standards on consent, data retention or responsible handling of these 'breakups' induced by updates.

Our reading is that this phenomenon will not disappear, nor should it be treated as a passing oddity: synthetic companionship will keep growing in a world with more loneliness, more isolation and more people seeking accessible, round-the-clock bonds. The relevant question is not whether AI can accompany —it already does, and it will keep doing so ever better—, but who audits these platforms, what protections exist for minors and what happens to the most intimate data ever shared by a person with an entity that owes them nothing in return. In the long term, an AI capable of sustaining emotionally sophisticated conversations is also the same technology that can relieve chronic loneliness, support mental health and free up human time for the relationships that do matter; but reaching that point with guarantees first requires seriously resolving the regulatory vacuum that this study exposes.