Are AI Companions Safe for Kids? What a Year-Long Study of 2,149 Adults Found
Folk & Dunn · Psychological Science, 2026 · Topics: Phones & Screens
⚠ General information only — this page explains one study and isn't medical or parenting advice for your specific child.
TL;DR: Here is the honest problem with this question: the best long-term study of AI companionship didn't study children. Folk and Dunn followed 2,149 adults in four countries for a full year and found a loop — people who felt more emotionally isolated went on to use chatbots for company more, and heavier use then predicted feeling even more isolated four months later. But the loop only fully appeared on one of their two loneliness measures, the authors themselves call the analysis exploratory, and not one participant was under 18. What the study gives parents is the shape of a risk, measured in grown-ups. What it can't give anyone is a verdict on kids.
The study followed adults, not kids
If you came here for a yes or no about your child, here is the honest version first. No long-term study has yet followed children or teenagers who use AI companions to see what happens to them over time. The strongest evidence that exists — the study on this page — followed adults.
That doesn't make it useless to you. Adults are where a new technology's patterns show up first, and the pattern this study found is exactly the one parents worry about: not a dramatic harm, but a quiet loop between loneliness and the chatbot that promises to fix it. It is worth reading as an early warning. Just remember the gap it has to cross: these were adults followed for a year, while the child a parent is worrying about is 13.
What did the researchers actually do?
Folk and Dunn ran a 12-month longitudinal study — the same people, surveyed again and again, rather than a one-off snapshot. Their participants were 2,149 adults across the UK, the US, Canada and Australia, surveyed every four months across a year; 979 of them completed all four rounds.
Each round, people reported how much they had used chatbots for social purposes — not ordering food or checking facts, but company: everyday conversation, talking through their day, asking for advice on a real decision, like whether to take a promotion. Just over a quarter had used a chatbot socially at least a few times in the previous four months. People also reported how lonely they felt, measured two different ways — and that detail turns out to matter, so hold onto it.
Because the same people answered repeatedly, the researchers could ask the question a snapshot never can: which moves first? Does loneliness send people to chatbots, do chatbots leave people lonelier — or both?
Is Character AI safe for kids? Why that's the name parents search
Most parents don't search for "AI companions." They search for a name — usually Character.AI, the app their child already has, or a friend has. A Common Sense Media survey from July 2025 found that 72% of teens have used an AI companion at least once, and about 1 in 3 said they had discussed something serious with one. Common Sense Media's own recommendation is that under-18s shouldn't use AI companions at all, and in November 2025 Character.AI removed open-ended chat for under-18s.
Those facts and this study come from different places, and they need to stay apart. The survey is about teenagers and real products. The study is about adults and a behaviour — "social chatbot use," one broad self-reported category that lumps a quick chat and a deep attachment together. Folk and Dunn never tested Character.AI, or Replika, or any named app, and they never studied a single teenager. So the study can't stamp any product safe or unsafe. What it can do is show what happened, over a year, when grown-ups leaned on chatbots for company — and let you judge how much of that pattern could apply to a much younger person.
What the study found: a loop that runs both ways
On one of the study's two loneliness measures — a single question about feeling emotionally isolated — the answer came back as a loop. When someone reported feeling more emotionally isolated at one check-in, they went on to use chatbots for social purposes more over the next four months. Feeling lonely pushed people toward the bots.
Then it flipped. People who had ramped up their social chatbot use felt even more emotionally isolated four months later. Loneliness led to the chatbot, and the chatbot led to more loneliness. Around and around, in the same people, across a full year.
Why might that be? The researchers' thinking — offered as an explanation to test, not a proven mechanism — is that chatbot company is easy in a way human company isn't. It is always awake, always kind, never busy, never needs anything back. The support feels real enough to reach for in the moment. But it may not be rewarding in the way that actually relieves loneliness, so the person reaches again — and the harder, richer work of real relationships can start to feel like even more effort by comparison.
The researchers also measured loneliness a second way, through a broader lens: social connection — feeling part of things, having people around you. And on that measure, only half the loop survived. Feeling less connected still predicted turning to chatbots later. But chatbot use did not significantly predict losing connection afterwards. One measure says loop. The other says half a loop. Both results are in this article because both are in the paper.
What this study can't tell you
It studied adults — full stop. Every participant was an adult. Their friendships were already formed, their schooling was behind them, and they knew perfectly well what a chatbot is. A child's social world is still under construction. Whether an always-available friend makes the slower work of real friendship feel harder by comparison is a serious question this paper raises with real weight. It does not answer it, and neither can this page.
"Safe" here means one thing: loneliness. 'Safe' is a bundle of worries for most parents — what the app might say to a child, what it does with their data, whether it can steer them. Folk and Dunn measured none of those things. It followed loneliness and social connection only. Its authority covers that ground and no further.
It can't prove cause. This is an observational study. People weren't assigned chatbots; they chose them, and the people who choose chatbot company differ from those who don't. Something else — a hard year, a lost job, a shrinking social circle — could be driving both the loneliness and the use. The design can show what predicted what, in order, over time. It cannot isolate the chatbot as the cause.
The authors say be careful — in their own words. The analyses were exploratory, and Folk and Dunn urge caution in drawing strong conclusions. The links they found were real but modest in size. This is initial evidence about a new behaviour, published by researchers who would rather you held it lightly than turned it into a headline.
If your child already talks to one: what to watch for
The study can't make this decision for you, and nothing in this section is a finding — it's the pattern from the paper, translated into the questions a parent can actually use:
- Watch for replacement, not minutes. The risk pattern in this research isn't using a chatbot; it's what the chatbot quietly takes over. A bot a teenager uses to rehearse a hard conversation before having it with a person is a tool. A bot that has become the main event — the one who hears about the day first, the friend who is always awake and never argues — is the pattern this study followed into deeper isolation in adults.
- Ask what the serious talks are about. About 1 in 3 teens who use AI companions have discussed something serious with one (Common Sense Media, 2025). If your child is taking real problems to a chatbot, the question worth asking isn't only "is that safe?" — it's "who else could hear this, and how do we make that door easier to open?"
- Know the product's own rules. Character.AI removed under-18 open-ended chat in November 2025, and Common Sense Media recommends no AI companions for under-18s at all. If your child is on a companion app, it's worth knowing whether they're inside the age the product itself now allows — and that the app being open to them is not the same as evidence about what it does to them.
- Keep the human doors easier to open than the app. Be the person who hears about the day. The adults in this study drifted toward the bots when they felt isolated; the protective move, at any age, is making sure a lonely kid's easiest available company isn't software.
Frequently Asked Questions
Is Character AI safe for 13 year olds?
No study can answer that yet — including this one. Folk and Dunn studied adults only; there was no 13-year-old, and no teenager of any age, in their sample, and they tested no product by name. What exists for a 13-year-old is guidance, not findings: Common Sense Media recommends no AI-companion use for under-18s, and Character.AI itself removed open-ended chat for under-18s in November 2025. The adult evidence on this page shows why researchers are paying attention — a loop between loneliness and chatbot company — but applying it to your child is a judgement call, and anyone who tells you a study has settled it is overstating the science.
Is Character AI safe for teens?
The honest answer has the same shape. Teen use is common — 72% of teens have used an AI companion, per Common Sense Media's July 2025 survey — but common isn't the same as studied. The only long-term evidence we have comes from adults, where heavier social chatbot use predicted feeling more emotionally isolated months later on one measure (and not significantly on another). Nothing followed teens over time. So "safe for teens" is currently unanswerable from research; what parents have is the adult pattern, the survey numbers, and the product rules — which is exactly why the watching brief in the section above matters more than a label.
Is Character AI safe to use?
For an adult looking at their own chatbot use, at least the people in the study match. Its answer: it depends what the use is doing. Light, occasional social use was common in the sample and isn't what the findings are about. The pattern to respect is the loop — reaching for the chatbot because you feel isolated, and feeling more isolated afterwards, on the study's emotional-isolation measure. And even that comes with the authors' own caution attached: the analyses were exploratory, the effects modest, and on the broader measure of social connection the second half of the loop didn't show up. Used as one tool among real relationships, nothing here sounds an alarm. Used as a replacement for them, this is the first long-term evidence that the replacement may not hold.
📄 The paper at a glance
Folk D, Dunn E. How Does Turning to AI for Companionship Predict Loneliness and Vice Versa? Psychological Science. 2026;37:276-286. DOI: 10.1177/09567976261427747
- Design: 12-month longitudinal panel — same people surveyed every 4 months; observational, so prediction over time, not proof of cause
- Who: 2,149 adults in the UK, US, Canada and Australia; 979 completed all four waves. No children or teenagers were studied
- Headline result: a two-way loop on the emotional-isolation measure — loneliness predicted more social chatbot use four months later, and more use predicted greater emotional isolation four months after that
- The split result: on the broader social-connection measure, only the first half held — feeling disconnected predicted later use, but use did not significantly predict losing connection
- What it doesn't show: anything about children or teens, any named app's safety, or the content/privacy side of "safe" — none of those were studied. The authors call their analyses exploratory and urge caution
The takeaway
Parents keep asking whether AI companions are safe for kids, and the truthful headline is uncomfortable: the research hasn't reached kids yet. What it has reached — 2,149 adults, followed for a year — is a loop worth knowing about. Loneliness sent people to chatbots, and leaning on chatbots predicted deeper loneliness behind it, at least on the measure that asks about feeling emotionally isolated. A companion that is always awake, always kind and never needs anything back sounds like a gift. Whether it can stand in for a real friendship, nobody knows yet. Until a study follows young people themselves, the parent's job is the one in the section above: watch what the chatbot is replacing.
Never miss a new article
New research, explained in plain English — on the Parenting by Science YouTube channel, with new videos every week.
Disclaimer: This article explains scientific research and is not medical or parenting advice for a specific child.