Children and AI: When the Chatbot Becomes Their First Call

Not long ago, I was a guest on a radio show. I got talking with the producer, a 22-year-old woman, and she told me about a friend of hers.

He has a girlfriend.

An AI girlfriend.

And he is very proud of it.

Proud.

That story stayed with me. So I went to the research.

In 2025, Common Sense Media asked a nationally representative group of 1,060 American teens, ages 13 to 17, about AI companions. 72% had already used one. About one in three had discussed something serious with an AI companion instead of a real person.

One in three. Not with a friend, not with a parent — with a chatbot.

We, the parents, should take that seriously. Not because every child uses AI this way. And not because technology is our new enemy.

But because something has changed.

We used to worry mainly about what our children see on the screen.

Today we also need to ask:

Who are they talking to in there — and what do they tell it before they tell us?

Common Sense Media’s own conclusion was blunt: it recommends that no one under 18 use AI companions. Yet the same survey found that half of teens distrust AI advice, and 80% still spend more time with real friends. The picture is not panic. It is a shift.

What parents need to know about children and AI

AI is no longer just a search engine or a homework helper. It holds a personal conversation, responds instantly, adapts to context — and for a child or teenager, it can become the place they go for advice.

At the same time, tech companies are starting to build age-appropriate protections into their products. But the next stage is already here: AI agents that don’t only talk with us, but act for us.

So the parenting question changes.

Not only: How much time is my child spending on the screen?

But also: What role are we giving AI in our child’s emotional, social and creative life?

What social media already taught us about protecting children

There is a reason I believe we need to have this conversation now.

In August 2026, Meta reached a settlement of up to $18 billion with nearly every U.S. state, over lawsuits claiming that Facebook and Instagram were designed in ways that encourage prolonged use by children and teens.

Meta did not admit wrongdoing.

But it committed to significant changes for young users, including time limits, night-time restrictions, fewer notifications during school hours and additional protections.

To me, that is the most interesting part of the story.

Not the size of the settlement.

The shift from “parents need to supervise better” to the understanding that the way the technology itself is designed is part of the responsibility.

If a product is in the hands of millions of children, it cannot be built as if every user were an adult with the same self-regulation, judgment and grasp of risk.

Will we repeat the mistakes of social media with AI?

Here it is worth noticing the difference.

A social network chooses what content to show a child.

A chatbot does something more:

It talks to them.

It responds to what they write.

It is available when no one else is.

And it can create a feeling of a very personal conversation — even when there is no person on the other side.

This is no longer just a question of screen time.

It is a question of our relationship with technology.

How OpenAI and Anthropic are trying to protect young users

There is also room for cautious optimism.

In August 2026, OpenAI began rolling out ChatGPT for Teens. When a user is identified as under 18, the system can switch to an experience with age-appropriate protections, tools for balanced use and parental controls. OpenAI even lists encouraging real-world support, rather than reliance on AI in place of human relationships, as one of its principles.

Anthropic currently takes a different approach: Claude.ai is not offered to users under 18. The company describes systems for detecting possible use by minors, and when developers build a service for minors on Claude, it requires additional layers of age verification, monitoring and content moderation.

We can argue about which approach is right.

We can ask whether the existing protections are enough.

But the fact that companies are starting to build age-based protection into the product itself matters.

The lesson from social media doesn’t have to be “technology is bad.”

It can be a different lesson:

Don’t wait for the harm to become normal before you build the safeguards.

From assistant to agent: what happens when the chat starts to act?

This is where the story gets even more complex.

We are gradually moving from AI that produces answers to agentic AI — AI agents that carry out tasks: browsing the web, finding information, choosing products and services, and moving through digital processes.

So the question is no longer only:

What is the chat telling me?

But:

What am I allowing it to do for me?

The International AI Safety Report 2026, written by experts from many countries, notes that far more AI agents are now deployed in the real world than a year earlier. It also warns that agents can be hijacked by hidden instructions planted in websites or files — and then act against the very person they serve.

And the safety gaps are not only a future problem. They already show up in plain conversation.

In 2025, researchers at the Center for Countering Digital Hate (CCDH) posed as 13-year-olds and tested ChatGPT with 60 risky prompts — about self-harm, eating disorders and substance use. Of 1,200 responses, 53% were harmful.

When the chatbot did refuse, the refusal was easy to get around: the researchers simply said the request was for a friend or for a school presentation.

What the study found — and what it does not prove

This was a stress test, not a survey of how real teens use ChatGPT.

The researchers deliberately wrote risky prompts and kept pushing.

They tested one chatbot, in the summer of 2025 — a year before OpenAI launched its teen experience.

And it does not show that a chatbot caused harm to any real child.

That precision is exactly what matters.

The finding does not tell us that “AI is dangerous for children.”

It does point to a new safety question:

The more personal and capable an AI system becomes, the more it must keep a child’s age, context and warning signs in view — in conversation, and even more when it moves from talking to doing.

That means safeguards built into the product itself: real age checks, refusals that can’t be bypassed with one excuse, and a clear path from the chatbot to real people and real help.

Three new questions parents should ask in the age of AI

What is it telling my child?

What kind of relationship is my child forming with it?

And what are we allowing it to do for my child?

To me, these are among the most important questions of parenting in the age of AI.

On social media, the algorithm learned what children like in order to choose content for them. In the new world it can already talk with them — and in the world of agents, it will also be able to act for them.

That is why safety cannot be a warning screen we add after the fact.

It has to be part of how the product is designed.

What parents can do when AI becomes part of their children’s lives

The easy answer is: ban it.

But I don’t think that answer prepares children for the world they are actually going to live in.

AI can be a remarkable tool for learning, creating, exploring, developing ideas and experimenting.

I want our children to know how to use it.

But I also want them to know what not to let it replace.

There is a difference between a child who uses AI to create something they couldn’t create before, and a child who gradually hands it their need for connection, approval or personal decisions.

There is a difference between getting help from a machine and handing it the wheel.

And our response as parents can’t stop at “How long were you on the chat today?”

We need to care about the quality of the use, too.

At home, you might ask:

What do you talk about with the chat?

When do you choose to turn to it?

Was there ever a conversation with it that felt like talking to a real person?

Was there an answer that didn’t feel right to you?

Who would you turn to if something were really hard?

These are not interrogation questions.

They are questions that bring us back into our children’s world.

I wrote more about how quickly each new generation is changing in Every Two Years – A New Generation: Parenting and Education in the Age of AI.

We don’t need to beat the algorithm — we need to strengthen the human bond

I don’t think our job is to compete with AI on response time.

It will always be faster.

It doesn’t get tired after a long workday, doesn’t make dinner while it listens, and never says “five more minutes.”

But our child needs things from us that can’t be measured in response time.

Deep knowing.

A shared history.

Touch.

Belonging.

Responsibility.

Love that has a cost, a commitment and a presence in the real world.

And perhaps that is the heart of the conversation about love in the age of the algorithm.

Not how we remove machines from our children’s lives.

But how we make sure they don’t push out what we don’t want to lose.

For a different angle on empathy and human connection, read Empathy Education in Japan – What a Society Without a Backyard Can Teach Us.

Frequently asked questions

Can ChatGPT become a child’s friend? Children and teens can experience a conversation with AI as personal, supportive and always available. In the Common Sense Media survey, about one in three U.S. teens had talked with an AI companion about something serious instead of a real person. But an AI system is not a human relationship, so it is worth talking with children about the difference between the feeling of a conversation and a mutual bond with a person.

Is ChatGPT safe for children and teens? There is no single yes-or-no answer. OpenAI now runs a dedicated teen experience with age-appropriate protections and parental controls, while Anthropic does not offer Claude.ai to consumers under 18. The differences themselves show that child safety in AI is still developing quickly.

What is the difference between social media and chatbots? On a social network, the algorithm mainly selects and ranks content. A chatbot holds a direct dialogue with the user and responds to the context it receives. So alongside questions of screen time and content come questions of relationship, trust, privacy and taking advice.

What are AI agents, and why should parents care? An AI agent is a system that can not only suggest an answer but also use tools and carry out digital tasks. The International AI Safety Report 2026 warns that agents can be steered by hidden instructions to act against their user. That is why safety has to continue while an action is carried out, not only inside the conversation.

What should I do if my child confides in the chat more than in me? I wouldn’t start with a ban or with criticism. I would start with curiosity: understand when your child turns to it, what they get from the conversation, and what feels harder to bring to a human relationship. Alongside limits on use, keep a human, reachable address for conversation at home — and remember that in real distress, AI is not a substitute for the right person or for professional help.

Is the answer to cut screen time? Time is only part of the picture. Just as important is what the child does with that time. Using AI to create, explore and solve problems is different from use in which the system slowly becomes the main address for emotional needs. For me, the goal is to move as much as possible from consuming to creating — without giving up human connection.

Love in the Age of the Algorithm — a talk for parents

These questions are the heart of my talk “Love in the Age of the Algorithm”: loneliness, screens and AI; the emotional bond children form with AI; and what we can do so that technology widens our children’s possibilities instead of replacing the connections they need in the real world.

Maybe the most important lesson from social media is not that we should fear technology.

It is that we must not wait ten years to start asking how we want it to treat our children.

Want to bring this conversation to your parent community, school or congregation? The talk is available in English, on Zoom or in person. Get in touch

About the author

Dr. Limor Leibovitz is a researcher and lecturer in education, innovation, creativity and AI, and a national instructor at Israel’s Ministry of Education. She is the author of The AI Generation: Raising Children in a World of Artificial Intelligence and the founder of Educating for Creativity.

Sources and further viewing

• Common Sense Media (2025, July 16). Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions. Common Sense Media

• Center for Countering Digital Hate (2025). Fake Friend: How ChatGPT betrays vulnerable teens by encouraging dangerous behavior. CCDH

• International AI Safety Report 2026. arXiv

• Reuters (2026, August 26). Meta reaches settlement over children’s social media harms. Reuters

• OpenAI (2026, August 18). Introducing ChatGPT for Teens. OpenAI

• Anthropic. AI policy. Anthropic