These findings do not mean that Nepalis generally prefer machines to people.
They do, however, show why the simple act of saying “I am not okay” can be complicated.
In families where emotional struggles may be interpreted as weakness, in communities where privacy can be difficult to maintain, or in situations where professional help is difficult to access, an anonymous-looking screen can feel easier to approach.
That is especially relevant for younger people.
UNICEF and WHO said in December 2025 that 5.2 percent of Nepali adolescents aged 13 to 19 had diagnosable mental health conditions, while 18.3 percent of children aged 6 to 18 reported emotional or behavioural difficulties. The organizations also said many affected children and young people receive little or no support and emphasized the need to reduce stigma and normalize help-seeking.
Being heard is not the same as being understood
There is an important distinction here.
An AI chatbot can respond immediately. It can help someone organize thoughts, put feelings into words or explore questions they are embarrassed to ask another person.
But a response is not the same thing as a relationship.
AI does not know a person's life in the way a trusted friend, parent, partner, teacher or mental health professional can. It does not see the person's face, hear the hesitation in their voice or necessarily understand everything happening outside the conversation.
And an AI response can be wrong.
A 2025 review of research on chatbots and mental health found evidence of potential benefits, but also emphasized that the field remains limited by methodological weaknesses and unanswered questions. A separate systematic review of generative-AI mental-health chatbots highlighted both their promise and their risks, including the need for stronger safety measures and more research in culturally diverse settings.
That distinction becomes particularly important when someone is dealing with a serious mental health crisis.
A chatbot may be a place to start a conversation. It should not automatically become the only place a person turns.
When the chat becomes the relationship
There is another question that is harder to answer.
What happens when someone does not simply use AI to find words, but begins to depend on it for emotional connection?
Research is starting to examine that possibility.
A four-week randomized controlled study involving 981 participants, conducted by researchers from the MIT Media Lab and OpenAI, examined loneliness, real-world social interaction, emotional dependence on AI and problematic AI use. The researchers reported that people who voluntarily used the chatbot more tended to show worse outcomes across these measures, while also emphasizing that the findings should not be interpreted as proof that chatbot use itself necessarily causes those outcomes.
That caution matters.
A lonely person may use AI more because they are already lonely. In that case, loneliness could be part of the reason for heavy use rather than simply its consequence.
The bigger question is therefore not whether AI makes people lonely.
It is whether a technology designed to be constantly available can sometimes become a substitute for relationships that are difficult, imperfect and necessary.
Human relationships require patience. They require vulnerability. Sometimes they involve misunderstanding and disappointment.
AI can remove much of that friction.
That can make it easier to talk.
It can also make it easier to avoid the harder conversation.
The uncomfortable question
Perhaps the most important question is not:
Why are people talking to AI?
It is:
Why do some people feel they cannot talk to anyone else?
If someone tells a chatbot about their fear before telling their closest friend, that does not necessarily mean the chatbot has won.
It may mean the friend was never made to feel safe enough.
If someone can type “I am falling apart” at midnight but cannot say those words at breakfast, perhaps the problem is not simply the technology. Perhaps it is the absence of spaces where vulnerability is accepted without embarrassment.
Nepal has been trying to address that gap. The government launched a National Mental Health Campaign in 2025 with support from WHO and UNICEF, with goals including reducing stigma, encouraging help-seeking, increasing family involvement and making people more aware of where support is available.
But campaigns and services can only go so far.
Listening begins much earlier.
It begins when a parent does not immediately say, “Everyone has problems.”
It begins when a friend does not turn another person's pain into gossip.
It begins when a partner listens without immediately trying to win an argument.
It begins when someone can say, “I don't know what is wrong with me,” and hear, “You can tell me.”
What the AI confession is really telling us
There is something revealing about the sentence: I told AI because I couldn't tell anyone else.
The remarkable part is not necessarily that a person talked to a machine.
The remarkable part is that they had something to say and nowhere they felt safe saying it.
AI may become a temporary listener, a place to rehearse difficult words or a first step toward understanding what someone is feeling.
But if the conversation ends there, society may miss the larger message.
People do not only need answers.
They need people they can talk to.
They need relationships in which honesty does not immediately become shame.
They need families where saying “I am struggling” does not feel like failure.
And they need access to professional help when the problem is bigger than what a friend or a chatbot can handle.
Maybe the future is not about choosing between humans and AI.
Maybe the more important challenge is making sure that nobody has to choose a machine simply because speaking to another human feels impossible.
Because sometimes, when someone says, “I told AI because I couldn't tell anyone else,” the story is not really about the machine.
It is about everyone else who was not there when the person finally decided to speak.