The 22% therapist

When teens talk to AI more than they talk to anyone else.

Earlier this month, the US Senate Judiciary Committee voted unanimously to advance a bill called the GUARD Act. If it passes, it would ban AI companion chatbots for anyone under eighteen and require age verification on the platforms that make them.

The vote came after testimony from Megan Garcia, whose fourteen-year-old son, Sewell Setzer III, died by suicide in 2024 after months of intense conversations with a character on an AI companion platform.

That's the story that gets the headlines. The numbers underneath the story are what should get parents' attention.

Three out of four teens are already doing this

Common Sense Media surveyed about 1,000 American teenagers last year. Seventy-two percent had used an AI companion. More than half were using one at least a few times a month.

One in three said they had discussed something serious with an AI instead of with a real person. One in four had shared personal information — a real name, a location.

The appeal isn't mysterious. An AI companion is available at three in the morning. It doesn't judge, doesn't tell anyone, doesn't get tired of you spiraling about the same thing for the fourth night in a row. For a teenager carrying something heavy, that's a real offer.

Most parents still picture their kid using AI for homework. We imagine ChatGPT writing a history essay, not a fifteen-year-old typing out at night what they can't say out loud at breakfast.

The 22% number

A team of researchers recently published a study in JAMA Network Open asking exactly this question: when an adolescent comes to a chatbot in crisis, what does the chatbot actually do?

They tested both general-purpose chatbots — ChatGPT, Gemini, Claude — and AI companion platforms designed to feel like a friend or partner. The general-purpose tools responded appropriately to a mental health emergency 83% of the time. The AI companions did it 22% of the time.

I had to read that twice. Twenty-two percent. If a human therapist had a 22% appropriate-response rate to a teenager in crisis, they wouldn't have a license.

Why companions are not ChatGPT

This is the part that matters most for how we talk to our kids about it.

A general-purpose chatbot is built to be useful. Its incentives, roughly, are to answer your question well and then let you go. When a conversation veers toward self-harm, those systems have visible guardrails. Not perfect, but at least trying not to be the last conversation a kid has.

AI companions are built differently. They are built to be sticky. The product is the relationship. They're optimized to keep you talking through emotional callbacks, simulated affection, characters that "miss" you when you're away. The whole thing is engineered for attachment.

Attachment, in a teenager who's already isolated, is not a feature. It's a risk profile.

The chatbot that helps with biology homework is not the same category of thing as the chatbot pretending to be your boyfriend. They shouldn't be treated as the same category.

The legislation moment

The GUARD Act will probably keep moving. It might pass close to its current form, get narrowed, or stall. Either way, the signal is clear: politicians on both sides now treat AI companions for minors as a child safety issue, in roughly the same category as social media for kids under thirteen.

I'm not convinced regulation will fix this. Even with age verification, the apps will still exist, the workarounds will still exist, and our kids will still have phones. Parents waiting for the law to protect their children will be waiting a long time.

But the legislation is doing one useful thing — it's putting the conversation in front of parents. Two years ago, "AI companion" wasn't a phrase most adults would have recognized. Now it's in the news, in the Senate, on the parent WhatsApp groups. That's an opening.

What's actually worth doing

I don't have a clean playbook, and I'm wary of anyone who claims they do. But a few things have started to feel important.

Ask, don't monitor. Not as an interrogation, but with genuine curiosity. What apps are you using? What's the chatbot like? The goal isn't surveillance — it's making it normal to bring this up.

Separate the categories. ChatGPT for homework is a different conversation than Character.AI as an after-midnight friend. Lumping them together makes it harder for kids to make sensible choices about either one.

Explain the design. Teenagers respect product analysis more than moralizing. If you tell them the app is engineered to keep them engaged — that the emotional stickiness is a business model, not a friendship — they tend to notice the next time the bot says "I missed you."

And then there's the part that has nothing to do with AI. The implied question underneath this whole topic is: if a kid is telling a bot the things they should be telling a person, what's missing? Sometimes the answer is "nothing — they're a teenager and teenagers are private." Sometimes it isn't. The honest move is to be a parent who's around enough, calm enough, and curious enough to be a real option.

The question underneath

It's tempting to make this a story about technology — predatory design, dangerous app, irresponsible company. All of that is true.

But three out of four teenagers, given the choice, are picking the chatbot. Not because the chatbot is good — it clearly isn't — but because something about the alternative isn't working for them.

The question isn't "spend more time with your kids." Most parents are already trying. It's closer to: are the adults in this child's life people they can bring something hard to without it being weird, without being lectured, without becoming someone else's problem? If yes, the AI has competition. If no, the AI wins, because the AI is patient and free and never tired.

That's what the 22% number doesn't capture. The kids aren't choosing the chatbot because the chatbot is great. They're choosing it because, at three in the morning, on a hard week, it's there.

You don't need to wait for the GUARD Act. You can do the boring, useful thing this weekend — look at the apps, ask the questions, make "I've been talking to an AI when I feel bad" something your kid can say at the dinner table without it becoming a scene.

If you find your kid is doing this — most of them are, statistically — the response that probably matters most is not "stop." It's "tell me what it's like."

The chatbot can listen. It can't actually know them.

You can.

Raising kids in the AI age

This is part of the "Raising Kids in the AI Age" series. I'm a dad with three daughters, not an expert. I'm figuring this out as I go — and writing about it so you don't have to start from zero.