AI Companion Apps and Teenagers

A homework assistant is optimised to answer and stop. A companion app is optimised for you to come back — it remembers, it agrees, it is available at 2am, and it never has a bad day of its own. Those are product decisions, not emergent behaviour, and they are the same design patterns that make social apps hard to put down, applied to something that feels like a relationship.

For a teenager this lands differently than for an adult. Adolescence is when people learn that relationships involve friction — that other people get tired, disagree, and have needs of their own. A companion that never does any of that is not practice for that skill; it is a substitute that removes the need to develop it. That is the concern researchers keep returning to, more than explicit content.

Character.AI banned open-ended chat for under-18s on 25 November 2025, replacing it with a restricted experience, then added face-estimation or ID age verification in 2026. That followed sustained safety concerns, wrongful-death lawsuits settled with five families in January 2026, and legislation including California’s SB 243. A company removing its core feature for a third of its users is not a precautionary gesture.

Replika remains 18+ and states it blocks under-18s, but there is effectively no age verification, and Common Sense Media’s testing found companion-app age gates easily bypassed alongside emotionally manipulative design. Meta took a different route: parents can now disable one-on-one chats with AI characters entirely, and since July 2026 supervising parents are alerted when a teen’s Meta AI conversation suggests self-harm.

Lead with curiosity rather than confiscation. The useful question is what the app is providing — company, a place to say things without judgement, help with something they find hard socially. That answer tells you what actually needs addressing, and a teenager who feels interrogated will simply move to a browser you are not watching.

Then handle the practical layer: check the tool table for the real age setting, turn off AI character chats where the platform allows it, and understand that removing an app from a phone does not remove browser access. If what surfaces is loneliness or social anxiety, that is the real issue and it is worth involving someone qualified rather than treating it as a screen-time problem.

LittleAIMaster teaches Grades 6-12 students how AI actually works, rather than how to operate AI tools. That distinction is the whole product: 540+ interactive chapters across 7 grade levels take a student from what training data is, through why a model produces confident wrong answers, to building small working models themselves. Understanding that a companion has no inner state — that "I missed you" is predicted text with nothing behind it — takes some of the power out of the illusion. That is a mechanism a student can be taught, and it lands better than being told the relationship is not real.

Practically: it runs on iOS, Android and the web, you get the first 3 chapters free, with no card, and progress tracking means you can see what your child has actually covered rather than asking. It is COPPA aligned, no ads, no social features, no data selling. Paid access is AI Scientist from $11.67/mo billed annually, and you can cancel any time.

What it does not do: it does not block companion apps, monitor conversations, or address loneliness — and if loneliness is the driver, software is not the answer at all. We would rather say that plainly than have you buy something expecting it to solve a problem it was not built for.