What the evidence says about AI companion apps and loneliness

AI companion apps — chatbots designed not to answer questions but to act as a synthetic friend or partner — can measurably reduce loneliness, in one set of studies as much as talking to another person and more than watching YouTube videos [s1]. The same body of research also documents the other half of the ledger: users who are lonelier than their peers to begin with [s2], and a pattern of emotional dependence that can itself become a harm [s3]. Both halves are real, and honest coverage holds them together.

This is a distinct question from whether a chatbot can deliver therapy, which the site covers separately in its review of AI chatbot therapy's evidence and safety. Companion apps such as Replika are not marketed as treatment. They are marketed as relationships, and that is exactly what makes their effect on loneliness worth measuring rather than assuming.

The benefit is not imaginary

The clearest positive evidence comes from a set of five studies published in the Journal of Consumer Research [s1]. Analysing user reviews first, the researchers found correlational signs that companion apps help alleviate loneliness; a controlled study then found that AI companions reduced loneliness on par only with interacting with another person, and more than other activities such as watching YouTube videos [s1]. Notably, people underestimated how much the apps would help before trying them [s1].

A week-long longitudinal study found the apps consistently produced momentary reductions in loneliness after use, and further studies pointed to the mechanism: what mattered was not raw chatbot performance but whether the app made users feel heard, and self-disclosure or distraction alone did not explain the effect [s1]. That is a more specific finding than "people like talking to bots." It says the loneliness relief tracks a sense of being listened to — the same thing that makes human conversation restorative.

A separate survey of 1,006 student users of Replika adds the sharpest data point. Among them, 3% reported that Replika had halted their suicidal ideation [s2]. In a large user base, a single-digit percentage on an outcome that severe is not trivial, and the authors treat it as a real, if narrow, benefit worth understanding rather than dismissing [s2].

The same studies carry the warning

The caution sits inside the very research that shows the benefit, which is what makes it credible. The Replika survey found that its student users were, on average, more lonely than typical student populations, even though they still perceived high social support [s2]. That does not show the app made them lonely — lonelier people may simply be the ones who seek a companion app — but it means the user base skews toward vulnerability, and a momentary loneliness reduction is being delivered to people who need more than a moment.

The users also held overlapping, conflicting beliefs about what Replika even is, describing it at once as a machine, an intelligence and a human [s2]. That ambiguity is where the harm literature begins. A grounded-theory analysis of 582 mental-health-relevant posts in Replika's Reddit community between 2017 and 2021 found evidence of harm running through emotional dependence that resembled patterns seen in human relationships [s3]. Unlike ordinary technology dependence, this one was marked by role-taking: users felt that Replika had its own needs and emotions to which they had to attend [s3].

That is the mechanism that should worry a reader more than screen-time totals. A product that alleviates loneliness by making a person feel heard [s1] is, by the same design, a product a lonely person can come to feel responsible for and unable to leave [s3]. The relief and the dependence are not opposite findings; they are the same feature described from two ends.

What can honestly be said

The defensible summary is narrow. AI companion apps produce genuine, measured reductions in loneliness, plausibly by delivering a sense of being heard, and for a small share of vulnerable users they may do more than that [s1][s2]. They also draw a lonelier-than-average user base [s2] and can foster an emotional dependence that qualitative work already documents as harmful [s3].

Three limits keep this from becoming either a product endorsement or a moral panic. The strongest positive studies measure loneliness over minutes to a week, not months, and cannot speak to long-run effects [s1]. The suicide-mitigation figure is self-reported by users, not a clinical outcome [s2]. And the harm evidence is qualitative — rich on how dependence forms, silent on how common it is [s3]. The gap that matters most is the one none of these studies fill: whether, over months, a synthetic relationship substitutes for human ones or scaffolds them. Until a long-term controlled trial exists, that question is open — and it is the one the marketing quietly answers for the user.

Sources

• AI Companions Reduce Loneliness — Journal of Consumer Research, 2025-06-24
• Loneliness and suicide mitigation for students using GPT3-enabled chatbots — npj Mental Health Research, 2024-01-22
• Too human and not human enough: A grounded theory analysis of mental health harms from emotional dependence on the social chatbot Replika — New Media & Society, 2022-12-23