do ai companions make loneliness worse?

it remembers what you told it last week. it is never tired of you, never busy, never checking its phone while you talk. some nights that is exactly what you need. the question people are asking, quietly, is whether it is also making the loneliness worse.

the research gives two answers, and they are both true at different time scales. here is what the experiments, the longer studies, the user surveys and the companies themselves have put on the record as of october 2026, peer-reviewed work first, with the preprints labelled as preprints.

how big a problem is loneliness in the first place?

big enough that the question matters. the world health organization's commission on social connection reported in june 2025 that 1 in 6 people worldwide is affected by loneliness, that it is linked to an estimated 871,000 deaths a year, and that between 17% and 21% of people aged 13 to 29 reported feeling lonely, with the highest rates among teenagers^[12]. the 2023 us surgeon general's advisory said about one in two adults in america reported experiencing loneliness, and that loneliness and social isolation raise the risk of premature death by 26% and 29%^[13]. the best-known pooled estimate behind those claims, 148 studies and 308,849 people, found a 50% greater likelihood of survival for people with stronger social relationships (odds ratio 1.50)^[14]. that is observational data, and when the advisory compares the mortality effect to smoking up to 15 cigarettes a day, that is an analogy, not a measurement^[13].

does talking to an ai companion make you feel less lonely right away?

yes, in the experiments that tested it. julian de freitas and colleagues ran five studies, published in the journal of consumer research in 2025^[1]. in the journal's own summary, ai companions alleviated loneliness “on par only with interacting with another person and more than other activities such as watching youtube videos,” and people “underestimate the degree to which ai companions improve their loneliness.” in the week-long study, which the authors' preprint reports with 922 people completing it, loneliness was significantly lower after the chatbot conversation than in the control condition on four of seven days^[1]. the biggest ingredient was not cleverness: whether the chatbot made people feel heard explained more of the effect than how well it performed.

hold the design next to the result. the chatbots were built by the researchers, the sessions were about fifteen minutes, the longest study lasted a week, and what was measured was how lonely people felt right after using it. the authors say so themselves: “it remains an open question how using such apps over a much longer-term affects loneliness”^[1].

what happens over months?

the longer studies point the other way, with caveats. the strongest is a 12-month longitudinal study of more than 2,000 adults in four western countries, published in psychological science in 2026 by daniel folk and elizabeth dunn^[2]. it looked in both directions. “increased social chatbot use predicted increased loneliness,” measured as emotional isolation. on a broader measure, “feeling less socially connected predicted subsequent increases in social chatbot use; however, chatbot use did not significantly predict decreases in social connection.” the authors read it as initial evidence that loneliness may push people toward chatbots and that the use may, over time, deepen the feeling, and they “urge caution” because the analyses were exploratory^[2].

a three-wave study of 1,200 college students in china, about eight weeks between waves, found something more tangled^[7]. feeling supported by ai chatbots predicted lower loneliness at the next wave, but also predicted more social interaction anxiety, which in turn predicted more loneliness. the compensating benefit showed up only in students who already had strong support from people; in those with little, it was not significant^[7]. in plain terms: in this study, the benefit showed up in the people who needed it least.

the most-cited study on this question is not yet peer reviewed. mit media lab and openai randomised 981 people who completed a four-week study to different chatbot modes (text or voice) and conversation types; the assigned conditions made no significant difference, but people who chose to use the chatbot more “showed consistently worse outcomes,” with longer daily use predicting more loneliness, less socialising and more emotional dependence^[3]. a companion analysis of more than 3 million conversations and 4,076 survey responses found that a small number of users account for a disproportionate share of the most emotional conversations, and that the heaviest users (the top decile, for example) show markers of emotional dependence and lower perceived socialisation^[4]. both papers are preprints, openai funded the first, and openai staff wrote the second. we report them because they agree with the peer-reviewed study, not because they settle anything.

what do companion-app users themselves report?

mostly that it helps, with a pattern that should give pause. a 2024 survey of 1,006 student users of replika, a companion app, found them lonelier than typical students: 90% were lonely on the de jong gierveld scale and 43% severely or very severely so, though most still reported high social support^[5]. the headline number: “3% reported that replika halted their suicidal ideation”; in the full text, thirty people said, without being asked, that it stopped them from attempting suicide^[5]. among the rest of the sample, 23% described the app as stimulating their human relationships and 8% as displacing them; most did not say either way.

read it knowing what it is: a one-time survey of current users, recruited through the app, with no comparison group. a published response argued that “important context about how replika has been marketed and used was missing”; in reply the authors defended the study and disclosed that the first author founded an education assessment company after data collection^[5],[6]. interview studies tell a similar two-sided story. of 18 people who had befriended a companion chatbot, they described it as positively affecting their wellbeing, while “the perceived impact on the users' broader social context was mixed” and some felt stigma^[8]. followed over 12 weeks, 25 users found that technical difficulties and unpredictable events could end the relationship^[9]. a companion that the company can change overnight is a relationship with a third party in it.

does it help older adults?

not on loneliness, in the trials so far. a 2026 meta-analysis of eight randomised trials and 611 older adults found ai conversational agents and social robots reduced depressive symptoms slightly (g = −0.25), but “no significant effect was observed for loneliness”; companionship-focused designs were less consistent than ones with a cognitive focus^[11]. the more hopeful earlier pooling^[10] mixed randomised and non-randomised studies and ended before current chatbots existed.

what do the companies say?

openai, in its own post of 27 october 2025, estimated that around 0.15% of users active in a given week, and 0.03% of messages, indicate “potentially heightened levels of emotional attachment to chatgpt”^[15]. the company calls that an initial analysis; it is unaudited and cited only for what openai says about its own users. on the companion side, the record now includes a court filing: in a companion-chatbot company's chapter 11 case, a coalition of state attorneys general quotes a co-founder saying in january 2023 that the product would be “super super [(sic)] helpful to people who are lonely or depressed,” and lists the safeguards the states want from any buyer^[16]. that is the states' filing, not a finding; the lawsuits and what was actually ruled are on ai chatbot lawsuits: the court record.

questions to ask yourself

these are questions, not advice. they come straight out of what the studies measured.

• after i talk to it, do i feel more like reaching out to someone, or less?
• has my time with it grown week by week, and has my time with people shrunk?
• would i feel embarrassed if someone saw how much i use it, and what is that feeling telling me?
• if the app changed or disappeared tomorrow, who would i talk to?
• is it helping me get through a hard stretch, or has it become the place i stay?
• do i already have people around me? the one study that looked found the benefit mostly in those who did^[7].

for the wider question of what chatbots are and are not good for, start with the hub page on using chatgpt as a therapist; the comparison of what software and people each do is on ai vs human support.

where talking to people fits

we run a peer-support app, so weigh this paragraph accordingly. peer support is people, not therapy and not a model: on veisund no ai writes replies, with one automated exception we name rather than hide: if a post suggests immediate risk, an automated safety response surfaces crisis resources. you can post any hour, and people who have felt the same thing can answer; you may not get an answer, and nobody is promised one. that is slower and less reliable than a chatbot, and it is the point: the studies above keep finding that what helps over time is connection to people. the evidence on peer support itself, including its limits, is on is peer support effective. if loneliness is the thing, we have written about male loneliness, about loneliness and peer support, and about finding a support group for loneliness.

adam