- Tension: A pilot study found that an AI dating simulator reduced loneliness and distress in chronically single men — but left their hostility, rigid gender beliefs, and tendency to blame women completely unchanged.
- Noise: The conversation around AI companions oscillates between techno-utopian enthusiasm and moral panic, missing the nuance that the technology in this study was designed as a therapeutic access point, not a replacement for human connection.
- Direct Message: The emotional wounds of chronic loneliness and the belief systems that grow around them require fundamentally different interventions — and feeling better is not the same as thinking differently.
To learn more about our editorial approach, explore The Direct Message methodology.
Thirty-two men walked into a university lab in Montreal, sat down at computers, and started flirting with a chatbot named Marie. They weren’t there for fun. They were there because they hadn’t been in a romantic relationship for at least a year — many much longer — and a team of researchers led by sexology professor David Lafortune at the Université du Québec à Montréal wanted to know whether practicing romance with artificial intelligence could shift something real inside them. According to new findings reported by PsyPost, it did — and it didn’t — in ways that reveal far more about modern loneliness than any dating app ever could.
The study, published in Archives of Sexual Behavior, used a custom-built platform called Kindling — designed to mimic the feel of real dating apps — and paired each participant with a text-based AI companion named Marie for a single structured session lasting less than two hours. The session broke into three 15-minute stages: initiating contact, practicing self-disclosure and vulnerability, and — critically — experiencing romantic rejection in a controlled environment. Between each stage, a licensed psychotherapist sat with participants to process what they’d felt.
The results, tracked at follow-ups, showed that participants reported sustained decreases in loneliness, general mental distress, and sexual distress. Several participants explicitly noted that the exercise was “highly enjoyable” and helped them “practice flirting without the usual fear of embarrassment,” that it allowed them to “lower their guard and express themselves more freely than they normally would.”
And then there’s what the intervention didn’t change.
Participants’ baseline hostility remained the same. Their rigid gender beliefs stayed intact. Their tendency to blame women for romantic failures — unchanged. A two-hour session with an AI and a therapist, it turns out, can soften the emotional bruising of chronic singlehood without touching the cognitive architecture underneath it. That gap — between feeling better and thinking differently — is where this story gets interesting.
Involuntary singlehood, as public health researchers describe it, affects a growing segment of the population: people who deeply desire intimate relationships but face persistent barriers to forming them. The consequences aren’t abstract. Research has linked chronic social isolation to higher rates of depression, anxiety, and cardiovascular conditions. The men in Lafortune’s study weren’t simply lonely in the casual, Sunday-afternoon sense — they were experiencing a form of social deprivation that compounds over time, eroding confidence, narrowing social behavior, and reinforcing the very avoidance patterns that keep them isolated.
What makes this study notable isn’t just the AI component — it’s the population it targeted. As Lafortune’s research motivations make clear, many men dealing with involuntary singlehood avoid traditional mental health care entirely. Sometimes this stems from deep distrust of therapists, financial constraints, or a general discomfort with discussing intimate insecurities. The AI dating simulator was designed, in part, as a side door — a way to reach men who would never voluntarily sit across from a therapist and say the words I don’t know how to connect with someone.
The question of whether AI belongs in that therapeutic space is already generating friction. As I explored in a previous piece on the ethics of robot therapists, the boundary between tool and replacement collapses fast when the “patient” begins forming emotional attachments to the software. And the risk here is specific: research indicates that millions of people now interact with romantic conversational agents globally. Some psychology experts have cautioned that frequent, unsupervised use could blur the lines between reality and simulation, potentially increasing emotional dependency on AI rather than reducing the barriers to human connection.
Lafortune’s team was aware of this. The study embedded a human psychotherapist into every stage — the AI was the practice field, but the therapist was the coach. Participants didn’t just flirt with Marie and go home. They processed the feelings that surfaced. They talked about what rejection felt like — even simulated rejection — and what it reminded them of. The design suggests the researchers understood something important: the AI isn’t the therapy. The AI is the thing that makes the therapy accessible.
But the study’s limitations are significant, and the researchers were transparent about them. The sample size was small. There was no control group — meaning there’s no way to determine whether the improvements came from the AI interaction, the human therapeutic support, the act of showing up and trying, or some combination of all three. The demographic was narrow: educated, white, heterosexual Canadian men. A single session, no matter how well-structured, can’t establish long-term behavioral change. The follow-ups showed sustained improvement, but the study wasn’t designed to track whether those gains held at six months, a year, or the moment a real person said no.
There’s a psychological concept worth naming here — the idea that practicing a feared behavior in a low-stakes environment can reduce the anxiety associated with performing it in real life. Exposure therapy operates on a similar principle: graduated contact with the thing that frightens you, in conditions you can control, until the fear response softens. The Kindling platform attempted this with romantic interaction. The three-stage design — approach, vulnerability, rejection — mirrors the actual emotional arc of dating, compressed into a format where the worst outcome is a chatbot gently turning you down.
The fact that participants’ hostility and gender rigidity didn’t shift raises a harder question. Emotional relief and cognitive restructuring are different processes — you can feel less lonely without examining why you believe women owe you attention. You can gain confidence in flirting without interrogating the worldview that frames rejection as a personal attack. The study measured both, and the divergence in outcomes is its most revealing finding. What it suggests — carefully, provisionally, within the limits of a pilot study — is that the emotional wounds of chronic singlehood and the belief systems that sometimes accompany them may require fundamentally different interventions.
This matters beyond the lab because the population Lafortune was studying isn’t small, and it isn’t static. The discourse around involuntary singlehood has grown louder and more politically charged over the past several years. Online communities organized around romantic frustration have produced both genuine support networks and radicalization pipelines — spaces where legitimate pain gets weaponized into ideology. An intervention that reduces emotional distress without touching the ideological layer is useful, but it’s incomplete. The researchers seem to know this. Their work reads less like a solution and more like a proof of concept — a demonstration that the side door exists, and that some men will walk through it.
What’s quietly radical about this study isn’t the technology. It’s the framing. Lafortune’s team treated chronic singlehood not as a character flaw or a market failure but as a condition that produces real psychological harm — harm worth designing clinical interventions around. That reframe matters. When we treat loneliness as a personal shortcoming, the lonely person becomes responsible for their own rescue. When we treat it as a public health concern with measurable consequences, we start building infrastructure for support — even if that infrastructure is imperfect, even if it involves a chatbot named Marie and a therapist sitting in the next room.
Research on forgiveness at Harvard has found that replacing negative emotions toward others with more positive ones requires both a cognitive decision and sustained emotional work — that the two layers operate in tandem but aren’t interchangeable. The men in Lafortune’s study seem to have gotten the emotional shift without the cognitive one. They felt lighter. They didn’t think differently.
And maybe that’s where the real intervention begins — not with the AI, not with the simulation, but with the recognition that relationships require tolerating being misunderstood without converting that discomfort into blame. The technology can lower the drawbridge. The therapeutic work that follows — the kind that reshapes how someone sees themselves and the people they’re trying to love — still requires a human in the room. The pilot study proved the drawbridge works. What’s on the other side is the harder project, and it’s the one that no algorithm has figured out yet.
