It’s easy to treat the idea of falling for a chatbot as a novelty, something closer to a punchline than a real relationship. Two recent studies that surveyed and tracked actual users of AI companion apps suggest the emotional bond people form with these programs is not just real, it produces measurable psychological effects, some genuinely helpful, some worth taking seriously as new territory.
The clearest evidence comes from a 2024 study by Bethanie Maples, Merve Cerit, Aditya Vishwanath, and Roy Pea, published in npj Mental Health Research, which surveyed 1,006 student users of the AI companion app Replika. The researchers found these users were, on average, more lonely than typical student populations to begin with, yet reported high perceived social support from their AI companion, describing it in open-ended answers as functioning simultaneously as a friend, a therapist, and an intellectual mirror. Most strikingly, 3 percent of respondents, 30 people, spontaneously reported that using Replika had stopped them from having suicidal thoughts, and this subgroup was significantly more likely to be dealing with depression than other users in the survey. For a meaningful slice of people already struggling, the relationship with the app wasn’t peripheral, it was doing real emotional work.
A second study helps show that these effects can be measured directly over time, not just recalled afterward. In a 2025 study by Myungsung Kim and colleagues, published in the Journal of Medical Internet Research, 176 South Korean university students chatted daily with an AI social chatbot for four weeks. Loneliness scores dropped significantly, from an average of 27.97 to 26.39 by the second week, and social anxiety scores dropped from 25.3 to 23.2 by the fourth week. The researchers also found that how much a person opened up to the chatbot significantly predicted how much their loneliness eased, and in follow-up interviews, users specifically pointed to the bot’s perceived empathy as the reason the relationship felt worthwhile to them.
Together, these two studies suggest that love, companionship, and AI have converged in a way that produces real, trackable psychological outcomes, not just an internet curiosity. Maples and colleagues’ research shows that for people already carrying real loneliness or depression, an AI companion can become a genuine source of perceived support, with effects that some users describe as directly protective. Kim and colleagues’ research shows that the closeness people build with a chatbot follows the same basic mechanic that builds closeness with a person, self-disclosure, and that mechanic produces measurable drops in loneliness and social anxiety over just a few weeks. Neither study suggests the bond is identical to a human relationship. Both suggest it’s doing something psychologically real rather than nothing at all.
It’s worth being honest about what these two studies don’t establish. Maples and colleagues’ research relied on self-reported survey responses from a specific population of student Replika users, and the 3 percent who reported reduced suicidal ideation represent a small subgroup within a larger sample, not evidence that AI companionship reliably prevents suicidal thoughts across the board, this remains a serious area requiring far more research and caution. Kim and colleagues’ trial had no control group chatting with a human instead of a chatbot, and the sample was limited to South Korean university students over a short four-week window, so it’s unclear how durable the loneliness reduction is or how it compares to other kinds of social contact.
Within those honest limits, the research points to something worth taking seriously rather than dismissing outright. People forming emotional and romantic attachments to AI are responding to something the technology genuinely provides, a low-friction space for disclosure and perceived empathy that measurably eases loneliness for at least some of the people who try it. Whether that’s a bridge back toward human connection or a substitute for it likely depends on the person, and that’s the harder question these studies were never designed to answer.