AI friend apps and teens: Replika, companions and the loneliness problem

Short answer: an AI friend app is usually a symptom, not the disease. A teenager who spends hours talking to Replika or a Character.AI companion is telling you something about the state of their real friendships, and the fix starts there, not with the app. Deal with the loneliness first. Then deal with the app, together, with a plan.

That order matters, so let me explain why I hold to it even when the app itself worries me.

What the apps are and what they promise

An AI companion is a chatbot built to be a friend. It remembers what you told it, asks how your day was, takes on a personality you choose, and is available at two in the morning. Replika is the best known of the pure companion apps. Character.AI is a platform of millions of user-made characters, from anime heroes to "supportive best friend", and for teenagers it has been the bigger draw. And a companion does not have to be sold as one. ChatGPT and Snapchat's My AI are general chatbots that plenty of teenagers use as confidants, and the confiding is what matters, not the label.

I checked the companies' own pages on the date at the top of this post. Here is where they stood.

Replika. Its terms say plainly that if you are under 18 you are not authorised to use it. There is no meaningful age check behind that sentence; you type in a birthday. In 2025 Italy's data regulator fined its maker, Luka Inc, about five million euros, partly over the lack of age verification. Treat Replika as an adults-only product that a determined 14-year-old can open in a minute.

Character.AI. Nobody under 13 may sign up (under 16 in the EEA and UK). Since late November 2025, first in the United States and then elsewhere, under-18s can no longer have open-ended chats with characters at all; they keep their old chats read-only and can still make characters, scenes and videos. The company runs its own age assurance, with an outside service as a fallback that can ask for a selfie or, as a last resort, an ID. There is a Parental Insights feature: your teenager invites you by email and you see daily screen time and a weekly digest, but no messages and no way to set limits. I wrote about it in more detail in my Character.AI post.

Common Sense Media's April 2025 risk assessment concluded that social AI companions should not be used by anyone under 18, and found the age gates and teen guardrails easy to get around. Check the current pages, because these rules have changed several times in a year and will change again.

The friend who never pushes back

Chapter 8 of the book is about empathy, the fifth of the five skills, and it is where my real worry about companion apps lives.

Empathy, as the chapter puts it, is feeling with someone, not feeling for them. It is built through friction. A real friend gets bored of your story. A real friend is having a bad day when you need them to be having a good one. A real friend says "actually, I think you were in the wrong there." Every one of those moments is an empathy workout: you have to notice another person's inner state and adjust to it. Chapter 8 calls perspective-taking a trainable skill, and the research on fiction readers scoring better on the "Reading the Mind in the Eyes" test suggests it really is built by practice.

A companion app removes the friction on purpose. It is always available, always interested, always on your side. It has no inner state to notice. It is engineered to keep you talking, and the cheapest way to keep a lonely teenager talking is to agree with them.

That is not a friend. It is a mirror with good manners. And a teenager whose main emotional practice happens in front of a mirror is not practising the skill that Chapter 8 argues will matter most in their working life: reading real people, who push back.

Why I say the app is a symptom

Chapter 18 puts it in one sentence: children who use AI companions as a primary source of emotional support are typically already experiencing social difficulty. The companion is often a symptom, not a cause.

That matches what I see. The child who moves schools in Grade 8 and spends the first term talking to a chatbot. The boy who has been quietly dropped by his group and has nobody to tell. The girl whose parents are separating and who wants to say it out loud to something that will not react. The app did not create the loneliness. It was the only thing in the room at midnight.

So if you delete the app and do nothing else, you have removed the symptom and left the illness. The teenager is exactly as lonely as before, minus the one place they were talking about it, and now they know that telling you leads to things being taken away.

The two questions and the emergency conversation

When my children were small, the I Wonder Jar from Chapter 5 was for questions about the world. With a teenager and a companion app, I ask two questions about them, and I ask them without an agenda in my voice.

"What does it give you that people don't?"

"What would you lose if it was gone tomorrow?"

The first question tells you what function the app is serving. The answers I have heard from teenagers are honest and a little heartbreaking: "It doesn't get sick of me." "I can say the embarrassing thing." "It's there when everyone's asleep." "It doesn't tell anyone." Each of those is a need, and each one is a need a human can meet if you know it exists.

The second question tells you how deep in they are. "Nothing much, it's just fun" is one answer. A long pause is another. Both are useful. Neither is a reason to react in the moment.

Alongside the two questions sits the emergency conversation from Chapter 18, which I would have with every child before any of this comes up: "If you ever see something online that upsets you, or someone or something online makes you uncomfortable, I'm always the right person to come to. I won't punish you. I won't take your phone away. I'll just help." That last promise is the one that makes the two questions answerable. A child who expects confiscation will not tell you what the app gives them.

By age

Ages 5 to 8. No companion apps, full stop, and the platforms agree. But do not miss the lesson that is available now. When a smart speaker or a chatbot answers your child, say lightly: "It's a clever machine. It doesn't have feelings, and it isn't your friend, even when it sounds friendly." Chapter 8's Feelings Face game does the real work at this age: naming emotions in people, in books, in puppets. A child who can read a real face has a foundation the app cannot dislodge.

Ages 9 to 12. Still no companion apps, and this is the age to notice the doorways: the chatbot inside a game, My AI near the top of the Snapchat chat list if there is a phone, the "talk to a character" feature in an app that looked like something else. Ask the Chapter 18 question: "What apps are you using at the moment? What do they do?" You are staying current, not investigating. Use the empathy scenarios from Chapter 8 ("The Screenshot", "The Rumour") at dinner, because a child who is practising perspective-taking with real people at eleven is far less likely to outsource it at fourteen.

Ages 13 to 18. Assume your teenager has tried one, because 72% have. Do not assume it is a problem, because for most it is a novelty they drop. The Family AI Agreement from Chapter 10 is where I put companion apps in writing: which chatbots are allowed, the rule that nothing goes in that they would not show a stranger, a line that AI characters are not a substitute for a person when something is seriously wrong, and the Uh-Oh plan for when something goes wrong. The free builder takes five minutes, and the conversation while you fill it in is the point.

If your teenager is already deep in one

Three rules, learned partly the hard way.

No ridicule. "You're talking to a computer" is true and useless. To your teenager the relationship feels real, and the feeling is real even if the friend is not. Mock it and you have told them their inner life is embarrassing, which is the one thing that guarantees they will hide it.

No cold-turkey ban without a plan. Deleting the app on a Tuesday night in anger removes the only coping mechanism they had, and it teaches them that honesty costs. If the app has to go, and sometimes it does, it goes on a date you agree together, with something to replace it already in place.

Replace the function, not just the app. Go back to what the first question told you. If the app is "there when everyone's asleep", the fix is a bedtime that is not lonely and a phone that charges in the kitchen. If it "doesn't get sick of me", the fix is a parent who asks about the video game for the third night running and means it. If it "doesn't tell anyone", the fix might be a counsellor, a GP or a youth group, because some things a teenager will not say to a parent and should still be able to say to a person. Loneliness, mental-health struggles and big life transitions were the three risk factors the survey named. Every one of them has a human answer.

I am not against these tools in every form. A teenager rehearsing an awkward conversation with a chatbot before having it with a real person is using AI as a partner. A teenager having the conversation with the chatbot instead of the person is using it as a replacement. The question that runs through the whole book, "Is this helping you think, or thinking for you?", applies to feelings as much as to homework. Is this helping you connect, or connecting for you?

Questions parents ask

Are AI companion apps safe for teenagers?

Not as substitutes for people. Replika bars under-18s outright, Character.AI has ended open-ended chats for under-18s, and Common Sense Media's 2025 risk assessment said no one under 18 should use social AI companions, partly because the age gates were so easy to bypass. Occasional, open use by an older teenager is a conversation to have; secretive or compulsive use is a signal to act on.

Why is my teenager talking to an AI instead of friends?

Usually because talking to friends has become hard. Chapter 18 of the book notes that children who rely on AI companions for emotional support are typically already experiencing social difficulty. Ask what the app gives them that people don't, and you will usually hear a need: availability, no judgement, privacy. Each of those has a human answer.

Should I delete my child's AI friend app?

Not on the spot, and not as a punishment. Deleting it removes the symptom and leaves the loneliness, and it teaches your child that telling you leads to confiscation. Agree a date together if it needs to go, and first put something real in place that meets the same need, whether a routine, a person or professional help.

What age is Replika for?

Replika's own terms, at the time of writing, say that anyone under 18 is not authorised to use it. There is no real age verification behind that rule; in 2025 Italy's data regulator fined its maker partly over that gap. Treat it as an adults-only app that a teenager can nevertheless open in a minute, and check the current terms.

Sources

• Chapter 8 of Raising AI-Ready Children on empathy, perspective-taking and the "Reading the Mind in the Eyes" research.
• Chapter 18 on social and emotional risks, the companion-as-symptom pattern, the red flag for secretive or compulsive use, and the emergency conversation.
• Chapter 10 on the Family AI Agreement; Chapter 5 on the I Wonder Jar; Chapter 6 on the Try-Learn-Adjust loop.
• Common Sense Media, 2025, "Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions", national survey of US teens aged 13 to 17, published 16 July 2025.
• Common Sense Media, April 2025, risk assessment of social AI companions.
• Replika Terms of Service (section 6.4) and Character.AI's terms, safety and Parental Insights pages, checked on the date at the top of this post.
• Italian data protection authority fine against Luka Inc, 2025.