Why are young people turning to ChatGPT instead of their peers when struggling?
- No judgment. The AI does not look at them differently, does not gossip, does not remember
- Available at 2am. No scheduling, no waiting, no worrying about being a burden
- Full control. They can stop mid-sentence, delete the conversation, never have to face the person again
- No consequences. What they say does not travel. It does not reach a parent, a teacher or a peer network
- Are we losing young people to a tool that cannot actually hold them, that can mimic empathy but offers no real relationship, no follow-up, no safety net if things get serious?
- Does the anonymity that makes ChatGPT feel safe also mean that the young people most at risk are the least visible to anyone who could help?
- Or is the real problem not that they are choosing AI, but that we have built peer and counselling spaces that feel less safe, less private, and less available than a chatbot?
Hi Rhoda, this is really an interesting one and Maybe the deeper reason is not simply that young people trust ChatGPT more than people, it may be that ChatGPT allows them to be vulnerable without having to become vulnerable in front of another human being. With a person, opening up can change a relationship: “Will they see me differently tomorrow? Will they tell someone? Will they think I’m weak?” With AI, they can say the things they are not yet ready to say out loud, test their feelings, ask questions they are embarrassed to ask, and even change their mind or leave the conversation without consequences. In that sense, ChatGPT may be serving as a private rehearsal space for vulnerability, a place where a young person can first understand and name what they are experiencing before they are ready to let another human into that part of their life.
@Rhoda Nakhosi This is a remarkably insightful, empathetic, and urgent observation. The reality you are describing is showing up across clinical settings, youth programs, and educational spaces globally.
Here is a structured, compelling response you can use to contribute to the discussion framing the trend not simply as a crisis, but as a critical "signal" for those working in youth wellbeing.
@Nonvicks Ochieng, you have laid out the mechanics of why AI feels safer very clearly. I want to press on one item in that list: "zero social stakes." That is true in the moment of the conversation, but is it true afterwards? A young person who has disclosed something painful to an AI is still carrying that disclosure alone. The AI does not gossip, but it also does not sit with them the next morning. So the social stakes have not disappeared, they have been deferred, and possibly compounded, because now, the young person has named something difficult to a machine and still has no human who knows. Do you see that deferral as a genuine risk, or as simply the cost of a trade off that is still net-positive for most young people?
@Rhoda Nakhosi The deferral of social stakes is a genuine risk that tips from a helpful trade-off into harm whenever AI becomes a terminal destination rather than a rehearsal. Externalizing pain to an algorithm can lower the barrier to eventually confiding in a human, but it creates a false sense of resolution. Crucially, a machine cannot provide somatic co-regulation or physical support the next morning. If disclosure ends with the screen, the young person remains completely alone in their real world compounding their isolation by replacing authentic, messy human vulnerability with a synthetic comfort that leaves their actual life unchanged.
@Mmbone Jessicah Christine, your phrase "private rehearsal space for vulnerability" has stayed with me, but I want to sit with the word rehearsal for a moment. A rehearsal implies a performance that comes later. The young person practices the disclosure, then eventually delivers it to a real person. If that is what is happening, then ChatGPT is functioning as a bridge, and our job is to make sure the bridge actually leads somewhere. But I am not sure that is always the case. For some young people, the AI may not be a rehearsal for human connection; it may be a substitute for it, and the relief of never having to perform the disclosure at all becomes the point. If that's true, then the concern is not that the rehearsal is happening in the wrong place. It is that for some young people, the rehearsal never ends. How would we even tell the difference between those two groups, given that both look identical from the outside?
@Mmbone Jessicah Christine Young people are increasingly turning to AI like ChatGPT during moments of struggle because it removes the social friction and emotional risks that often come with opening up to peers. Sharing personal vulnerabilities with friends carries the real fear of judgment, altered social dynamics, or gossip, whereas AI provides an environment with zero social stakes where one can express embarrassing thoughts or deep anxieties without shame. Additionally, young people often hesitate to confide in friends out of a desire not to burden them with their own heavy emotions, whereas an AI has infinite patience, never gets overwhelmed, and is immediately available 24/7 especially during late-night anxiety spikes when peers are asleep. Unlike human interactions, which can sometimes turn into shared panic or unsolicited drama, AI acts as a neutral soundboard that helps process chaotic thoughts into clear, structured steps, making it an accessible, low-risk outlet when human connection feels intimidating or unavailable.
What concerns me most about this case is not only the number 36, but what happens to those 36 girls after they become a statistic.
A school can report pregnancies. A health facility can report teenage mothers. The police can report cases. The education system can report re-entry. Yet none of these numbers, by themselves, tells us whether an individual girl is actually safe, supported and still in school six months later.
This is where I think we need to change the way we conceptualise adolescent pregnancy. It should not be treated exclusively as a reproductive-health indicator. It is also potentially an education, safeguarding, poverty, gender, social-protection and health-systems indicator. A sudden concentration of cases in one school should therefore trigger a coordinated vulnerability assessment rather than simply a debate about boarding facilities or punishment.
I would also be cautious about interpreting the absence of police reports as evidence that families are unwilling to act. Non-reporting itself needs investigation. Families may face stigma, fear, financial dependence, pressure to settle matters privately, limited knowledge of reporting pathways, or lack of confidence in formal institutions. The question should therefore be: what prevents a child-protection concern from moving from community recognition to formal protection and appropriate follow-up?
And there is another issue we should not overlook: the girls who are not in the data. The 36 pregnancies became visible because they occurred within one identifiable institution. How many adolescent pregnancies, sexual-abuse concerns, school dropouts or vulnerable girls elsewhere never reach any formal dataset?
For me, the real test of our systems is therefore not whether we can produce an accurate number of pregnancies. It is whether that number activates a response.
We should be able to follow a learner from identification → protection → health care → psychosocial support → school re-entry → retention → completion. If we cannot tell where a learner disappears along that pathway, then we have a surveillance system that counts events but a service system that loses people.
The proposed dormitory may have a role if distance and unsafe travel are demonstrated contributors, but infrastructure should be matched to the evidence. Otherwise, we risk solving the most visible problem rather than the most important one.
Ultimately, these girls should not become a short-lived news story, a school statistic or another entry on a county dashboard. The measure of success should be what happens to each learner after identification.
That is where I believe the conversation needs to move: from “How many girls became pregnant?” to “What did our systems do when we found them—and who made sure they were not lost afterwards?”
@Henry Magoba, I am going to do something slightly unfair and pull you into this thread, because your last post is about a different topic but the underlying question is the same one we are circling here: what happens after the number becomes visible? You argued that a count only matters if it activates a response, and that we should be able to follow an individual from identification through to completion. I would put the same question to this thread: when a young person discloses distress to ChatGPT, we have no identification, no pathway, no follow-up. By your own test, that is a system that counts nothing and loses everyone it touches. So I would be curious, does your "identification → protection → support → completion" framework have any purchase here at all, or is the AI case genuinely outside it because there is no point of entry for a system to respond to?
Thank you Rhoda for sitting with this uncomfortable truth. I have been sitting with it too since your workshop note, and I want to offer a different lens.
@Charles Bawate Anyama, this is the fullest answer anyone has given to my original question. But I want to test the parenting argument, because I think it lets our institutions off slightly too easily.
Most young people were not raised to open up about being mentally exhausted due to phrases like "You are a child, why are you struggling?" and the likes therefore they tend to hide how they are actually feeling when talking to peers or counsellors about their mental struggles because of fear of vulnerability. Some also grew up knowing that it is "western" or "rich" people culture. Lack of information about the relevant channels to speak to is also a major cause I'd say. If you ask most young people to mention a counselling centre outside of their school or places of worship, they might not be able to which might raise the question, are we really "losing them" if they are not aware of this channels? These are some of the reasons I would say contributes to them turning to ChatGPT as it tells them what they want to hear and gives them a sense of comfort needed at the time.
@Bleige Mutua, welcome so much! I really like this perspective, "If a young person cannot name a single counselling centre outside their school or place of worship, then we cannot really say they "chose" ChatGPT over us," you cannot choose what you do not know exists. That reframes the problem to part of what looks like a preference for AI may simply be a visibility failure on our side. It also makes me wonder whether the answer is not just better services but literally putting them where young people already are, with the same ease of access as opening an app. Thank you for adding this. It is a quieter explanation than the trust argument, but possibly a bigger one.
Rhoda, thank you for pushing — you are right. My first lens was parenting, and it did let institutions off the hook. Let me correct that.
@Charles Bawate Bawate Anyama, thank you for taking the push seriously and for answering directly. What stands out to me is your finding that anonymity alone was not enough, that the bridge only holds when someone says "can I check on you tomorrow morning?" That is a concrete, testable insight and it sharpens my original question, the thing that converts a rehearsal into a relationship is continuity, and continuity is exactly what a chatbot cannot fake. I also appreciate you naming the design failure honestly, 9-5, report-driven, visibility oriented services were built for us, not for a 15-year-old at 2am. That admission is worth more than any defence of the system.