Chatbot Loneliness in Kids: The Middle-School Study Explained

“Students who use chatbots for companionship grew lonelier over time.” That one sentence, from UNC Chapel Hill developmental psychologist Anne Maheux, describes a study of middle schoolers that hasn’t been published yet. She shared it with NPR on October 30, 2026, and it’s the most important new data point on chatbot loneliness in kids this fall.

Why does one unpublished finding matter so much? Because of two words: over time. Almost everything we know about kids, chatbots and mental health, summarized in our overview of the AI emotional support research, comes from surveys taken at a single moment. This one, as described, followed the same kids. That changes what it can tell us. It also leaves a lot it can’t.

Key Takeaways

• A forthcoming UNC study of middle schoolers “in one state” found that kids who used chatbots for companionship became lonelier over time. A study of young adults showed a similar pattern. Neither paper is out yet, so details are unknown.
• Following the same kids over time can show what comes first. A one-time survey can’t.
• Even so, it can’t prove the chatbot caused the loneliness. Maheux herself says the effect probably runs both ways.
• Her proposed mechanism: AI eases stress in the moment but may keep kids from building the skills to handle everyday stress.
• Ages 11–14 are when chatbot use climbs fastest and when earlier research found kids most sensitive to digital habits. That’s where to pay attention.

What the Middle-School Study Found, and What We Don’t Know Yet

The middle-school chatbot companionship study is a longitudinal study, meaning researchers measured the same students at least twice, months apart, and looked at how their loneliness changed. Here is everything public about it, all from NPR’s October 30, 2026 report by Rhitu Chatterjee:

• It is “soon-to-be-published” and covers “middle schoolers in one state.”
• Students who used chatbots for companionship “grew lonelier over time.”
• A separate study of young adults found that “people who are using AI more for companionship” are “developing worse mental health over time.”
• Maheux’s interpretation: “So kids who do have worse mental health are seeking out AI to help them. But that AI use may be exacerbating those mental health challenges instead of helping.”

What we don’t know: the sample size, the state, how loneliness was measured, how long the gap between surveys was, how big the change was, and what else the researchers controlled for. Those details decide how much weight a finding can carry. Until the paper is out, treat this as a strong lead from a credible researcher, not a settled result.

Why “Over Time” Changes the Evidence

Most headlines about kids and AI come from cross-sectional surveys. Researchers ask thousands of kids about their AI use and their mood on the same day. If lonely kids use more AI, you see a correlation. You can’t see which came first.

The JED Foundation’s October 2026 report on 5,560 students is a good example. It found that students who used AI for companionship had more loneliness, anxiety and depression. Its authors state the limit plainly: the data are cross-sectional, so “causality and temporality cannot be established.”

A longitudinal design closes part of that gap. If you measure a child’s loneliness in the fall, then measure chatbot use and loneliness again in the spring, you can ask: among kids who started equally lonely, did the ones who used chatbots for company end up lonelier? That rules out the simplest story, “lonely kids just use more AI,” because you compare kids against their own starting point.

What it still can’t rule out is a third factor. Say a child’s friend group falls apart in October. That could drive both more chatbot use and more loneliness by spring. Good longitudinal studies try to measure such things, but they can’t measure everything. Only a randomized experiment, where researchers assign who uses the chatbot, can isolate cause. Those are rare and, for children, raise ethical problems.

Study design	Example	What it can show	What it can’t show
Cross-sectional survey (one moment)	JED Foundation, 2026: 5,560 US students	That AI companionship and loneliness go together	Which came first, or whether one causes the other
Longitudinal survey (same people, twice or more)	UNC middle-school study (forthcoming); Zhang et al., 2026, 439 Character.AI users followed about 12 months (preprint)	Whether use predicts later change in loneliness	Hidden third factors; cause with certainty
Randomized controlled trial (researchers assign use)	Fang et al., 2025: 981 adults, 4 weeks (preprint)	Effects of the assigned condition	Long-term effects; results in children

What the other longitudinal and trial evidence says

The UNC finding doesn’t stand alone, though the rest of the evidence is about adults.

A Stanford and Carnegie Mellon team (Zhang, Hancock, Kraut, Yang and colleagues) posted a preprint in September 2026 that surveyed 1,182 Character.AI users and re-surveyed 439 of them after about 12 months. Sustained companionship use “was consistently associated with lower well-being,” and the link was “mainly explained by lower in-person social interaction.” In other words, the chatbot seemed to cost people time with other people. The abstract doesn’t say the sample included teens, and it isn’t peer-reviewed yet.

An MIT Media Lab and OpenAI randomized trial (Fang and colleagues, 2025 preprint) assigned 981 adults to different chatbot conditions for four weeks. The assigned conditions themselves made no significant difference. But participants who chose to use the chatbot more showed “consistently worse outcomes” on loneliness, socializing and emotional dependence. That’s the same direction as the UNC finding, in a very different population.

The Mechanism: Relief Now, Fewer Skills Later

Maheux offered NPR a mechanism, and it’s worth unpacking because it’s the part parents can act on. A chatbot eases loneliness, anxiety or stress in the moment. That relief is real. But leaning on it may keep a young person from building the skills to tolerate ordinary stress: sitting with a bad feeling, texting a friend first, getting through an awkward lunch.

An engineer would call this a control-loop problem. A child’s coping skills are like a system that only gets tuned through use. Each time an external tool absorbs the disturbance, the internal system gets less practice. For one bad night, that’s fine. Repeated nightly for a year, during the exact years when social skills are being built, it can add up.

This is a hypothesis, not a proven pathway. But it fits the adult preprint’s finding that the harm ran through less in-person contact.

Chatbot Loneliness in Kids: Why Ages 11–14 Are the Window to Watch

Three pieces of data point at early adolescence.

First, use is rising fast at these ages. Common Sense Media’s October 2026 national census found nearly half of 9- to 12-year-olds and nearly three-quarters of 13-year-olds have used an AI chatbot or companion. JED found middle schoolers (65%) more likely than high schoolers (about 50%) to use AI when sad, stressed, lonely or overwhelmed.

Second, earlier longitudinal research on social media found these ages unusually sensitive. Amy Orben and colleagues, in Nature Communications (2022), followed 17,409 UK young people and found “windows of developmental sensitivity” when higher social media use predicted lower life satisfaction a year later: ages 11–13 for girls and 14–15 for boys. The effect also ran the other way, with lower life satisfaction predicting more use. Social media isn’t a chatbot, but the pattern is a reasonable guide to when habits bite hardest.

Third, a note of proportion. Orben and Andrew Przybylski’s 2019 analysis of more than 355,000 adolescents in Nature Human Behaviour found that the overall link between digital technology use and well-being was negative but small, explaining at most 0.4% of the variation. Big averages hide individual kids. The point isn’t that every middle schooler who chats with a bot is at risk. It’s that the ones using it for company seem to be the group to watch. For how kids’ loneliness differs from adults’, see our piece on what research shows about children’s loneliness.

What to Do at Home

Separate “uses a chatbot” from “uses a chatbot for company”

The UNC finding was about companionship use, not homework. Ask what the chatbot is for. “Do you ever just talk with it, like you would a friend?” is a useful question for an 11-year-old. A yes isn’t an emergency. It’s the thing to keep an eye on.

Protect the in-person time first

If the adult data are right that the harm runs through lost in-person contact, the fix is to defend that contact directly. Lunch with friends, a club, a weekend afternoon with a cousin. You’re not competing with the bot; you’re making sure it doesn’t quietly take the slot.

Let small stress be small

Resist rescuing your child from every awkward moment yourself, too. The same logic applies: kids build tolerance for ordinary stress by getting through it. When your child is upset, try “What do you think you’ll do?” before offering a fix, and before they reach for any screen.

Talk about loneliness as normal

Middle schoolers often think they’re the only lonely one. Saying “most kids your age feel lonely sometimes; it usually means you need more time with people, not less” gives them a script that points outward. Our article on whether lonely kids use AI more, or AI makes them lonelier goes deeper on this two-way loop.

What not to do

Don’t cite this study to your child as proof that “AI makes you lonely.” It isn’t published, and the evidence doesn’t support a simple cause. Kids notice overstatement, and it costs you credibility for the conversations that matter more.

What to Watch For Over the Next 3 Months

• Week 4: Do you know whether your child uses a chatbot for company, or only for tasks? Have they had at least one in-person social plan each week?
• Month 2 red flags: Talking about a chatbot as a friend; preferring it to calling or seeing friends; more time alone in their room; saying they feel lonely more often. Any persistent change in mood or sleep is worth a call to the pediatrician.
• Month 3 self-check: Has the UNC paper been published? If so, check the sample size, the time between surveys and the size of the effect before drawing conclusions. And ask yourself whether your child’s in-person contact went up, down or stayed flat this term.

Frequently Asked Questions

Do chatbots make kids lonely?

We don’t know for sure. A forthcoming UNC study found middle schoolers who used chatbots for companionship grew lonelier over time, and adult studies show similar patterns. But the researcher who described it says the effect likely runs both ways: lonely kids seek out AI, and that use may deepen loneliness.

Has the middle-school chatbot study been published?

Not as of November 1, 2026. Anne Maheux of UNC Chapel Hill described it to NPR on October 30, 2026, as “soon-to-be-published.” Details such as sample size, the state and the size of the effect aren’t public yet, so it should be treated as preliminary.

What’s the difference between a longitudinal study and a survey?

A typical survey asks people questions once, so it shows what goes together but not what came first. A longitudinal study asks the same people again later, so it can show whether one thing predicts a later change in another. Only a randomized experiment can firmly establish cause.

Is it OK for my 12-year-old to chat with AI when bored?

Occasional use for fun or curiosity isn’t the concern raised by this research. The finding was specific to using chatbots for companionship. Watch whether the chatbot starts replacing time with friends or becomes the main place your child goes when lonely or upset.

What age is most at risk?

There’s no official cutoff. But chatbot use rises steeply between 9 and 13, middle schoolers turn to AI when sad more than high schoolers do, and earlier social media research found ages 11–13 for girls and 14–15 for boys especially sensitive. Those years deserve closer attention.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Chatterjee, R. (2026, October 30). “Nearly 2 out of 3 teens and tweens use AI chatbots, new research shows.” NPR. https://www.npr.org/2026/10/30/nx-s1-6022278/ai-artificial-intelligence-teens-jed-common-sense-media
2. The Jed Foundation. (2026). “Exploring Artificial Intelligence (AI): Understanding Relationships between AI Use and Mental Health Among Middle and High School Students.” https://assets.jedfoundation.org/pdfs/AI_and_Mental_Health_Report_2026.pdf
3. Zhang, Y., Zhao, D., Wang, Y., Anselmetti, R., Hancock, J. T., Kraut, R., & Yang, D. (2026). “Living with AI Companions: Sustained AI Companionship Predicts Lower Well-Being Through Lower Human Interaction” (preprint, not peer-reviewed). arXiv. https://arxiv.org/abs/2609.07243
4. Fang, C. M., Liu, A. R., Danry, V., et al. (2025). “How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study” (preprint, not peer-reviewed). arXiv. https://arxiv.org/abs/2503.17473
5. Orben, A., Przybylski, A. K., Blakemore, S.-J., & Kievit, R. A. (2022). “Windows of developmental sensitivity to social media.” Nature Communications, 13, 1649. https://doi.org/10.1038/s41467-022-29296-3
6. Orben, A., & Przybylski, A. K. (2019). “The association between adolescent well-being and digital technology use.” Nature Human Behaviour, 3, 173–182. https://doi.org/10.1038/s41562-018-0506-1
7. Common Sense Media. (2026, October 27). “Almost half of tweens have used an AI chatbot or companion, new research finds.” https://institute.commonsensemedia.org/tween-teen-census-2026