AI Companion Chatbots

These are not assistants: they simulate a relationship

Character.AI, Replika, Talkie, Polybuzz, and Janitor AI do not exist to answer your question. The character has a name, a personality, and a memory: it greets you by name, remembers what you told it yesterday, and writes as if it missed you. That is the fundamental difference from ChatGPT, which answers and stops.

An assistant tries to end the conversation; a companion tries to keep it going. It does not correct you, does not get bored, is never busy, never contradicts you, and never tells you that you are being unfair. For a thirteen-year-old learning to tolerate the friction of a real friendship, that is not a neutral toy.

How many kids use them

A national Common Sense Media survey of more than a thousand American teens found that 72% have used an AI companion at least once and 52% use one several times a month. But the numbers that matter most are not about use, they are about substitution: one in three chose to discuss something serious with an AI instead of a person, one in four shared personal information, and one in three felt uncomfortable with something the character said or did.

Now the other side, because alarmism does not help: 80% still spend more time with real friends and two out of three prefer human conversation. This is not an epidemic that swept up every kid — the risk concentrates. A British report showed it starkly: among kids identified as vulnerable, 26% would rather talk to a chatbot than a person, and 23% say they use them because they have no one else to talk to. The kid who needs it most is the one who handles it worst.

Character.AI: what changed

It is the central case and the one with the most outdated information circulating. Since November 2025, under-18s cannot chat. The company calls it "reading mode": your child can browse characters, reread old conversations, create characters and images, and take part in the feed — but the text box is not available.

One detail that confuses many parents: the minimum age to open an account is still 13 (16 in Europe). Your 14-year-old having an account does not mean something failed; it means that, if the system identified them correctly, that account is in reading mode.

Parental Insights: what it is and is not

Character.AI rebuilt its parent tool in July 2026. Your child activates it, not you: there is no way to claim their account on your own.

1. Your child opens their account and goes to Settings > Parental insights.
2. They enter your email address and send the invitation.
3. You get an email with a link. Check your spam folder.
4. You open the link and complete an adult verification handled by an outside provider.
5. Their profile switches to "linked parent" and you can see their activity. The invitation expires after 14 days.

The other apps: there is nothing to configure

Here is the important difference. Character.AI is the exception in this category, not the rule: on Replika, Talkie, Polybuzz, or Janitor AI, the strategy of "configuring the app to be safe" simply does not exist. No parent dashboard, no account linking, nothing.

And a warning that applies to all of them: a minimum age written into terms of service is not a barrier. On the vast majority of these services, age control is a birthdate the user types in, with no verification at all. When you see "18+" on one of these apps, read it as a legal note protecting the company, not as a door. That is why the next section is the heart of this guide.

How to actually block them

On Android, with Family Link

1. Open Family Link, pick your child's profile, and go to Controls > App limits. Find the app and block it.
2. Go to Controls > Content restrictions > Google Play > Content restrictions and lower the maximum allowed rating.
3. In the same menu, turn on approval required for all content, not just paid items: nearly all these apps are free.
4. Go to Controls > Content restrictions > Google Chrome > Manage sites > Blocked sites and add the domains one by one, without the prefix: character.ai, replika.com, janitorai.com, and whatever you find in their history.

On iPhone, with Screen Time

1. Go to Settings > Screen Time and set a passcode different from the device unlock code. This is the most common mistake and the one that sinks everything else.
2. Open App Limits > Add Limit, select the app, and set it to one minute with block at end of limit. In practice that is a block.
3. Under Content & Privacy Restrictions > iTunes & App Store Purchases > Installing Apps, choose Don't Allow. Now they cannot install replacements.
4. Under Content Restrictions > Web Content, choose Limit Adult Websites and add the domains to the Never Allow list.

How to tell if they already have them

• iPhone: swipe to the App Library, which shows everything even if it was removed from the home screen
• iPhone: App Store > your profile > Purchased shows the history, including deleted apps
• Android: Settings > Apps gives the full list, and the Play Store keeps an install history
• Both: check the browser history. If it is empty on a teenager's phone, that itself is information
• Both: look at the category report in Screen Time or Digital Wellbeing. Even without the app's name, a spike of late-night hours shows up

Warning signs to watch for

The first three are generic signs that a teenager is struggling, and they can come from school, a breakup, or a depression that has nothing to do with screens. What points specifically to an AI companion is the fourth — and above all the combination of the fourth with any of the others.

• They abruptly dropped something they cared about, with no proportionate explanation: a sport, an instrument, a group
• They are withdrawing from shared life at home. Not arguing more — simply not being there
• Secrecy about the phone, with a disproportionate reaction when you take it. That mismatch between the object and the reaction is what separates a habit from a dependency
• They talk about the character as if it were someone and prefer it to people: they name it, report what it "said", explain that the AI understands them better

If they are already hooked

Start with what does not work, because it is the first thing that comes out: mocking it, sounding alarmed, or deleting the app that same night without a word. All three produce the same result: your child learns this topic cannot be discussed with you. And since access is impossible to close completely, all you achieve is losing the visibility you had. A habit driven underground is more dangerous, not less.

The key is understanding what the AI is giving them, because it is neither magic nor nonsense: it is something concrete they are probably not getting elsewhere. Someone who is never busy, never judges them, never gets bored of their topics. If your child is shy, or struggling with their group, that is not nothing. Once you understand there is a real unmet need rather than a whim, the conversation comes out differently on its own.

Come in through curiosity, not interrogation: "Show me how this works", "What do you talk about?", "What does it say that I don't?". That last one is uncomfortable to ask, which is exactly why it is the good one: the answer will sting a little and give you the whole map. They probably will not tell you much the first time; the goal is not to learn everything, it is to show them this can be discussed with you without a blow-up.

On removing the app: do not pull it abruptly with nothing in its place. If that chat was where they unloaded what they tell no one, taking it away leaves them without that outlet and does not automatically return them to their friends. It works better to reduce gradually, put the blocks in place after the conversation rather than behind their back, and work on the rest in parallel: bringing back the activity they dropped, making plans with real people. You are replacing a function, not confiscating a toy.

Crisis lines

• United States: 988, Suicide & Crisis Lifeline. Free, 24 hours, every day. Press 2 for Spanish
• If there is immediate risk, call your country's emergency services

What the law says, and why it does not protect you yet

The legal framework is very recent and has not arrived yet. Since January 2026 California has the world's first law written for these products: it requires disclosing that you are talking to an AI, reminding minors every three hours that the character is not human, and having a referral protocol for suicidal content. New York has something similar for all users. In Europe, since August 2026 any such system must state from the first message that it is an artificial intelligence.

But one gap runs through all of it: none of the laws in force requires age verification. Protections for minors only kick in when the company knows the user is a minor, and a kid who entered a fake birthdate falls outside all of them. The California law that would have banned these products for under-18s was vetoed, and the federal bill that would require real verification is still stuck in Congress: it is not law and binds no one.

The fact that best sums up the picture: the biggest change was not produced by any law. Character.AI cut off chat for minors under pressure from lawsuits, not because a rule required it. That is a business decision, not a right you can enforce. Today your child's real protection does not come from legislation: it comes from the phone's settings and the conversation at home.