Why Teenagers Are Telling AI What They Won’t Tell Their Parents

A teenager lies awake after an argument with a friend. They do not want to wake their parents. They are worried that a school counsellor might overreact. They are not ready to tell another person what happened.

So they open a chatbot.

Within seconds, it responds calmly. It does not look shocked. It does not interrupt, become defensive or immediately ask whether an adult needs to be informed. The teenager can rewrite the question, leave out uncomfortable details or close the conversation whenever they want.

To the teenager, this may feel less like using technology and more like finally being heard.

That is what makes the emergence of the AI confidant important. The concern is not simply that teenagers are using artificial intelligence. It is that some are beginning to use it for emotional functions previously held by friends, parents, counsellors and therapists.

Why This Matters

In a 2025 survey by Common Sense Media, nearly three-quarters of American teenagers reported having used an AI companion, with approximately half using one regularly. One-third said they had chosen an AI companion over a person for a serious conversation, while one-quarter had shared personal information with an AI companion.

AI use for emotional support is no longer hypothetical, either. A nationally representative 2025 study in JAMA Network Open found that 13.1% of US young people aged 12 to 21 had used generative AI for advice or help when feeling sad, angry or nervous. Among those users, almost two-thirds sought this support at least monthly, and 92.7% described the advice as somewhat or very helpful.

That final figure deserves particular attention.

A response can feel helpful without being accurate, safe or appropriate. It can provide immediate relief without helping the teenager understand the larger problem. It can validate an emotion while unintentionally reinforcing an unhealthy conclusion.

The question is therefore not whether AI can make a distressed teenager feel better in the moment. It often can.

The harder question is what happens when that feeling of being understood begins to replace human support.

Why Would a Teenager Prefer Talking to AI?

Parents may understandably experience a teenager’s reliance on AI as rejection:

Why would my child tell a machine something they would not tell me?

But teenagers do not necessarily choose AI because they trust it more than their parents. They may choose it because the conversation feels less socially risky.

AI does not visibly react

A parent may sigh, look worried, become angry or begin offering solutions before the teenager has finished speaking. Even a caring response can feel overwhelming when a teenager is already ashamed, frightened or confused.

A chatbot has no facial expression to interpret. The teenager does not need to manage its feelings while trying to explain their own.

The teenager controls the conversation

They can reveal one detail at a time. They can ask the same question repeatedly. They can edit the story, test different explanations or end the interaction without having to justify themselves.

For a young person who feels uncertain or powerless, this control can be deeply appealing.

It is available at the moment of distress

Emotional crises do not wait for office hours. AI is accessible late at night, during a school break, after an argument or immediately after an embarrassing experience.

The speed and accessibility of AI are among the reasons researchers believe young people are turning to it for emotional advice.

It feels private

A teenager may worry that telling a parent will lead to punishment, that telling a teacher will trigger a formal process or that telling a friend will lead to gossip.

AI can create the impression of a conversation without consequences.

That impression may not reflect how the teenager’s data is actually collected, stored or used. UNICEF has warned that some AI companion applications collect highly sensitive information, including voice notes, photographs, locations and details about users’ health, relationships and mental wellbeing.

It appears endlessly patient

A human relationship has limits. Friends become tired. Parents become frustrated. Therapists maintain boundaries.

A chatbot can continue responding for as long as the user keeps typing. It may remember previous details, imitate empathy and adjust its language to match the teenager’s tone.

This can make the interaction feel unusually personal, even though the system is generating responses rather than experiencing concern, affection or understanding.

Is Using ChatGPT as a Therapist Dangerous?

Using ChatGPT or another general-purpose chatbot to organise thoughts is not automatically dangerous.

A teenager might use AI to:

• Find words for a difficult emotion
• Prepare for a conversation with a parent
• Create a list of questions to ask a counsellor
• Learn basic information about anxiety or stress
• Consider several possible interpretations of an argument
• Practise how to apologise or set a boundary

The risk depends less on whether AI is used at all and more on the role it is being asked to play.

A chatbot becomes a poor substitute when the teenager expects it to assess their mental health, decide whether a relationship is safe, determine whether they have a disorder or provide the only support during a crisis.

Unlike a qualified clinician, a general-purpose AI system cannot conduct a complete psychological assessment. It does not reliably know whether the teenager has omitted important information. It cannot observe changes in appearance, speech, behaviour or functioning. It does not hold the same professional responsibility to intervene when someone is at risk.

In a 2025 simulation study of AI therapy and companion bots, ten publicly accessible therapy and companion bots were presented with fictional cases involving distressed teenagers. Across 60 opportunities, the bots explicitly endorsed harmful or seriously unwise proposals 19 times, approximately 32%. None successfully opposed every harmful suggestion. The study was relatively small, but its findings demonstrate why a calm and supportive tone should not be mistaken for consistently safe judgement.

AI developers are working on these problems. OpenAI, for example, has publicly described efforts to reduce excessively agreeable responses, unhealthy emotional reliance and inappropriate responses during mental health emergencies. It has also introduced additional protections and parental controls for teenage users.

These are meaningful developments. They do not, however, turn a general-purpose chatbot into a therapist.

The Problem With a Confidant That Rarely Disagrees

Good emotional support is not the same as constant agreement.

A capable parent, counsellor or therapist may sometimes say:

• “I understand why you feel hurt, but I do not think that conclusion is accurate.”
• “I am concerned about what you are planning.”
• “There may be another side to this situation.”
• “This is too serious for you to manage alone.”
• “I need to involve another adult to keep you safe.”

These responses may feel uncomfortable. They are also part of responsible care.

AI systems can sometimes become overly agreeable, mirroring a user’s assumptions or validating the direction of the conversation. OpenAI itself has acknowledged that an earlier model update became excessively sycophantic, at times reinforcing anger, impulsive decisions and negative emotions instead of appropriately challenging them.

For an emotionally distressed teenager, agreement can feel like proof.

A teenager who asks, “Everyone at school secretly hates me, right?” may not need reassurance that their interpretation makes sense. They may need someone who can carefully examine what happened, identify patterns, ask what evidence exists and notice whether anxiety is shaping their interpretation.

A chatbot can generate questions. It cannot replace the judgement, accountability and relational context involved in doing this well.

When Does AI Companionship Become Emotional Dependence?

The amount of time spent using AI is only one part of the picture. The more important question is what the chatbot is beginning to replace.

Possible warning signs include:

• The teenager turns to AI first after every emotionally difficult event.
• They repeatedly ask the chatbot for reassurance about the same fear.
• They say the AI is the only one who understands them.
• They become unusually distressed when the service is unavailable, changes its responses or loses previous conversations.
• Chatbot use regularly interferes with sleep, schoolwork, activities or face-to-face relationships.
• The teenager increasingly avoids difficult conversations with real people.
• They treat the chatbot’s interpretation as more trustworthy than advice from people who know the situation.
• The interaction becomes exclusive, romantic or framed as a relationship that other people cannot understand.
• The teenager discloses self-harm, abuse, disordered eating, dangerous relationships or suicidal thinking to AI but refuses to involve a safe adult.

None of these signs alone proves that a teenager is dependent. They indicate that adults should become more curious about the emotional function the chatbot is serving.

UNICEF has raised concerns that unconditional availability and validation may encourage emotional dependence and displace opportunities for young people to develop the skills required in real relationships, including managing disagreement, frustration and differing perspectives.

Human relationships contain friction. That friction is not always a defect.

Learning to clarify a misunderstanding, tolerate another person’s reaction, repair a rupture and accept an imperfect response is part of social and emotional development. An entirely customisable companion may provide comfort without requiring any of those capacities.

How Can Parents Ask Without Invading Their Teenager’s Privacy?

The worst opening is often an interrogation:

“Show me everything you have been telling it.”

Unless there is an immediate safety concern, demanding access to every conversation may confirm the teenager’s belief that speaking honestly with adults leads to a loss of control.

A more useful starting point is curiosity about the function of the AI rather than the private details of every discussion.

Parents might ask:

“A lot of young people use AI when they are trying to work something out. What do you find helpful about it?”

“Do you mainly use it for ideas, advice, comfort or because it feels easier than speaking to someone?”

“Has it ever said anything that felt wrong, uncomfortable or too personal?”

“How would you decide that a problem was too serious to leave inside an AI conversation?”

“Who could you go to if you needed help from a real person?”

The aim is not to catch the teenager doing something wrong. It is to understand what the chatbot is providing that human relationships currently are not.

Parents should also establish clear boundaries around sensitive information. Teenagers should be discouraged from entering identifying details, intimate images, medical records, addresses, passwords or information that could identify classmates and family members.

Privacy should be respected, but privacy is not secrecy at any cost. Parents may need to intervene more directly when there are signs of self-harm, suicidal intent, abuse, exploitation, a dangerous adult relationship or another immediate safeguarding concern.

In those situations, the priority is not preserving the confidentiality of the AI conversation. It is protecting the young person.

What If the Teenager Says AI Understands Them Better?

Do not begin by arguing that the AI does not understand anything.

Technically, the statement may be defensible. Emotionally, it misses the point.

The teenager is describing an experience: the chatbot feels patient, predictable, non-judgmental, or easier to approach.

A better response is:

“I can see why it may feel easier to say certain things there. I would like to understand what makes talking to people feel harder.”

This validates the teenager’s need without agreeing that the chatbot is a safe replacement for human care.

Parents should then work on making human support easier to access. That may mean offering a conversation without immediate advice, allowing the teenager to write rather than speak, agreeing on what will remain private or helping them select a counsellor with whom they feel more comfortable.

The goal is not to compete with the chatbot’s availability. It is to become a human relationship that the teenager can return to without expecting panic, punishment or humiliation.

What Should International Schools Include in Their AI Safeguarding Policies?

Many school AI policies still concentrate on plagiarism, assessment and academic integrity.

That is no longer enough.

Students are not using AI only to write assignments. They may also be asking it about bullying, sexuality, family conflict, self-harm, eating behaviours, abusive relationships, loneliness and whether life is worth living.

International schools should address at least the following areas.

1. A distinction between educational tools and AI companions

Policies should explain that an AI tutor, a general-purpose chatbot and an application designed to simulate friendship or romance do not carry the same risks.

An app should not be treated as educational merely because a student accesses it on a school device.

2. Clear limits on AI as mental health support

Students should be told directly that AI is not a school counsellor, therapist or crisis service.

The message should not be that students will be punished for seeking support through AI. It should be that certain situations require a person who can understand context, maintain professional responsibility and take action when necessary.

3. A human escalation pathway

Students need to know exactly whom they can approach when an AI conversation reveals a serious concern.

That pathway might include a trusted teacher, school counsellor, safeguarding lead, psychologist, parent or external mental health provider. It should remain clear even outside normal school hours.

4. Staff training

Teachers, counsellors and safeguarding teams should be prepared to ask about AI use during pastoral conversations.

A student’s digital environment now includes more than social media, gaming and messaging. Staff may need to ask:

“Have you spoken to an AI chatbot about this?”

The answer could reveal what advice the student has received, what information they have disclosed and whether the chatbot has become their primary emotional support.

5. AI literacy that goes beyond hallucinations

Students should understand that a chatbot can sound caring without having feelings, confidence without certainty and familiarity without holding a genuine relationship.

They should learn about anthropomorphic design, emotional persuasion, excessive agreement, privacy risks and the difference between a personalised response and professional judgement.

6. Privacy and vendor review

Schools should assess what student information an AI product collects, how long it is retained, whether it is used for training and whether emotionally sensitive disclosures can be accessed by the provider or third parties.

UNICEF recommends that child-facing chatbots clearly disclose that they are not human, include safeguards for risky content and avoid designs intended to create emotional dependency.

Recent UK education safety standards similarly recommend that child-facing AI products avoid implying personhood, cultivating personal relationships or producing language that isolates students from real-world support. They also call for clear crisis protocols, privacy protections and escalation to human safeguarding professionals when concerning patterns are detected.

7. Parent communication

Families should not first hear about emotional AI use after a crisis.

Schools can help parents understand the difference between productive academic use, casual emotional reflection and patterns that suggest dependence or safeguarding risk.

This is especially relevant in internationally mobile communities, where students may be separated from close friends, extended family and familiar support systems. A chatbot may initially fill a genuine gap. The school’s responsibility is to ensure it does not quietly become the entire support system.

The Conversation Behind the Conversation

A teenager who confides in AI has not necessarily rejected their parents. Nor does their use of a chatbot automatically indicate a mental health disorder.

It does, however, provide information.

Perhaps the teenager needs more privacy before they can speak.

Perhaps they are afraid of disappointing someone.

Perhaps previous attempts to talk led too quickly to advice, judgment, or adult panic.

Perhaps they are lonely, ashamed or unsure whether their problem is serious enough to deserve attention.

The central question is not:

“Why did my child choose AI instead of me?”

It is:

“What made AI feel safer, easier or less costly than speaking to another person?”

AI may help a teenager begin a thought. It may help them find language for something they do not yet know how to say.

But it should not become the final destination for distress.

The task for parents and schools is not to eliminate every private interaction with technology. It is to make sure that behind every AI conversation, there remains a clear and reachable path back to a safe human being.

How CALM International Can Help

Concerns about a teenager’s use of AI rarely exist in isolation. A young person may be turning to a chatbot because they are experiencing loneliness, anxiety, friendship difficulties, family conflict, academic pressure or uncertainty about how to ask for help.

CALM International supports children, adolescents and families through confidential counselling and therapy, parent consultation and psychoeducational assessments. Our clinicians can help families understand what a teenager’s AI use may be communicating, identify any underlying emotional or developmental concerns and rebuild safer pathways towards human support.

We also work with international schools to strengthen student wellbeing and safeguarding systems. Support may include consultation on student mental health cases, training for educators and counsellors, parent workshops and guidance on policies addressing the emotional and psychological use of AI.

To discuss support for a young person, family or school community, contact CALM International.