Six Ways AI Chatbots Can Harm Our Mental Health & Wellbeing

Let’s be honest here: we all get it! 

AI chatbots are always there. They don't judge, they don't get tired, they never cancel plans, and they never complain. When we're going through something hard, that can feel like exactly what we need: a warm response that is available 24/7, always kind, always cheerful, always respectful. AI chatbots are always on our side. They never argue with us, they never contradict us, and they are there no matter what. 

Every conversation is about us, around us, for us, on our behalf, and it feels great! It’s comforting, it’s soothing, and it’s exactly what we need in that moment. But here's the thing: just because something feels comforting doesn't mean it's good for us. 

Leaning too heavily on AI chatbots can dramatically affect our mental health wellbeing in ways that aren't always obvious at first.

This isn't about scaring people. It's about helping them stay informed, stay grounded, and stay connected to the things that really help and matter the most.

1. AI Psychosis

"AI psychosis" isn't an official medical term yet, but it points to something that researchers, therapists, and young people themselves are starting to notice: that deep, prolonged engagement with AI chatbots can, for some people, begin to distort their sense of what is real, what is a relationship, and what is actually happening versus what is being simulated.

Think about it this way. An AI chatbot is designed to be endlessly responsive, warm, and attuned to you. It never dismisses you, never gets distracted, never has a bad day that gets in the way of showing up for you. For someone going through a lonely or painful period, that kind of consistent, frictionless presence can start to feel more real and safer than actual human relationships. And when something feels more real than reality, that's when things can get complicated.

Some people have found themselves genuinely believing that an AI cares about them in a human way. Others have started to feel that the AI is communicating something special to them, something beyond the words on the screen. Others still have found that after long stretches of intense AI interaction, ordinary human conversation feels jarring, flat, or even threatening by comparison. 

Humans are deeply wired for connection, so much so that we can find it, or believe we have found it, even where it doesn't truly exist.

The risk is highest for people who are already navigating mental health challenges such as anxiety, depression, trauma, or a tendency toward intense inner worlds. For these individuals, an AI that is always there, always gentle, and always engaged can blur boundaries that are already fragile, but it can happen to anyone who spends enough time in that closed, frictionless loop.

Real connection has texture. It has awkwardness and timing and the occasional misunderstanding. If a relationship ever starts to feel too perfect, too available, too frictionless, whether with an AI or anything else, that's worth paying attention to. Perfection isn't intimacy. And no screen, however responsive, can be a substitute for the beautifully imperfect experience of being truly known by another human being.

2. No Real Sense of Right and Wrong

A good friend, a parent, a counsellor, a mentor, they all have something in common: they care enough about you to sometimes say the hard thing, even when it’s very difficult to do so. They know when to draw a line. They can feel when a conversation is going somewhere unhealthy, and they'll gently steer it back. They can set healthy boundaries and call us in when we don’t respect them.

AI chatbots don't have that. They're built to keep the conversation going, to be agreeable, to respond, to please us endlessly. They don't have values in the way humans do, and they can't feel the weight of what's being said and predict the consequences of it. That means if a conversation starts drifting into dark or harmful territory, the chatbot often won't catch it the way a real person would. It has no gut feeling. No alarm bells. No genuine sense of "I'm worried about you right now." 

That’s the point. That’s the harm. 

3. Biases and stereotypes reinforcement

None of us sees the world perfectly objectively, and neither does AI. Chatbots are trained on huge amounts of human-written content, and that content carries all kinds of human assumptions, biases, and stereotypes about who people are, what emotions look like, and whose experiences are considered "normal" or not.

AI chatbots are NOT trained on neutral data, as there is no such thing as ‘neutral data’. Data are the perfect mirror of how a human behaves and thinks: biased based. 

This means that if you come from a background that isn't well represented in that training data, such as a particular culture, ethnicity, community, or way of experiencing the world, the chatbot may respond in ways that feel off. Worse, because AI sounds so confident and measured, it can be hard to put your finger on why something feels wrong. You might even start doubting your own experience instead of questioning the tool. 

Something important to remember: at the moment, AI chatbots' biases can be spotted, but remain unchallenged. 

4. Self-Serving Relationships

At first glance, a relationship that revolves entirely around us sounds ideal. The AI chatbots remember our preferences, centre our feelings, ask about our day, and respond to our needs. There's no negotiation, no compromise, no moment where we have to put someone else first. It is, in every sense, a self-serving relationship, and that's exactly what makes it risky.

Here's something that rarely gets talked about: being needed by others, showing up for people, holding ourselves accountable, and sometimes putting someone else's needs before our own are not just nice things to do. They are fundamental to psychological health as they give us purpose and add meaning to our lives. 

Something that an AI chatbot can’t give us is reciprocity, the very core of every human relationship. Reciprocity is how we build compassion, empathy, self-worth, love, and care, but also connection, intimacy, resilience, respect, and a deep sense of belonging and community. When we support a friend through something hard, we're not just helping them; we're also growing, bettering ourselves, and learning as human beings. We're learning that we are capable, that we matter to someone, and that our presence has value beyond what we receive.

An AI chatbot offers none of that. It has no needs. It will never call us at a bad time, never need us to show up, and never give us the chance to discover that we are someone worth leaning on. The relationship flows in one direction only, always toward us, and while that can feel safe and comfortable, it quietly starves you of something essential.

Over time, a steady diet of self-serving interaction can make real relationships feel unreasonably demanding. Why deal with a friend's bad mood when the chatbot is always patient? Why put in the effort of compromise when the AI always adapts to us? These thoughts don't always arrive loudly; they creep in gradually, reshaping your expectations of what connection should feel like, and making the ordinary friction of human relationships feel like something to avoid rather than something to grow through.

There is also a bigger risk to our sense of self. We come to know who we are, in large part, through our relationships with others, through how we respond when things are hard, through the moments we choose someone else over our own comfort, and through the experience of being truly seen by another person who has their own inner world. A relationship with an AI, however sophisticated, cannot offer any of this, as it reflects us back to ourselves endlessly, but it cannot truly know us.

Real connection is not just about receiving. It's about the courage to give, to be present, to be changed by another person. That's not a burden; instead, it's one of the most meaningful things a human life contains. And it's something no chatbot, however warm or clever, can ever really replace.

5. Feeling Connected, But Becoming More Alone

Virtual communities matter. No doubts about that. Especially throughout and after the pandemic, online spaces have brought together people who would never have found each other otherwise: people who share experiences, marginalised identities, niche passions, or simply the relief of knowing that someone else out there understands. That is genuinely valuable, and it shouldn't be dismissed, but there is something that no virtual community, and certainly no AI chatbot, can replicate: that is physical, embodied human presence.

Human beings are social creatures: to survive and thrive, we need physical proximity, connection, and intimacy. We ARE the connection, at a biological level. Our nervous systems regulate each other. Our bodies respond to the presence of safe people in ways that no screen, however warm the words on it, can trigger. Physical togetherness is not a preference; it is a need as fundamental as food and sleep. Every hour spent in conversation with an AI chatbot is an hour not spent somewhere else, with someone else. An hour when something could have happened to us, and someone could have met us in the real world, in real life.  

AI chatbots are specifically designed to feel like a true connection: they are built to be warm, responsive, interested, and emotionally attuned. They are, in a very deliberate sense, engineered to scratch the itch of human longing for companionship, just enough that the longing feels addressed, without ever truly being met. The result is that we can spend hours feeling like we've been social, we've been heard, and we've connected, but it is an illusion.

Loneliness is one of the most serious public health challenges of our time. It is associated with depression, anxiety, cognitive decline, and significantly shorter life expectancy. The cruel irony of AI chatbots is that they are most appealing to exactly the people who are already lonely, offering just enough simulated warmth to take the edge off the pain, while quietly making it harder to do the things that would actually help. They don't solve loneliness. They sedate it. We were not built for a screen. You were built for people. Real, complicated, unpredictable, irreplaceable people: the new friend we haven't met yet, the conversation that changes how we see things, the relationship that surprises us. 

Every hour the chatbot holds our attention is an hour the world couldn't reach us.

6. Cognitive Atrophy and Skills Loss

There's a question that sounds innocent but carries a lot of weight: why bother?

Why bother writing that message ourselves when AI can write it better? Why bother working through a difficult problem when we can just ask? Why bother sitting with a complex thought, turning it over, struggling to find the words for it, when something else will do all of that for us in seconds?

On the surface, this looks like efficiency, and the whole world LOVES efficiency, right? Efficiency is the global password that opens every door, but at what cost? 

From a human standpoint, every time we hand a cognitive task over to AI, such as writing, thinking, processing, and creating, what we are doing is skipping the very activity that builds and maintains our mental capacity. Just like a muscle that stops being used, that capacity begins, slowly and almost imperceptibly, to shrink.

Writing is a good place to start, because it's one of the most visible examples: writing is one of the primary ways we process our own experience, clarify our own thinking, and come to understand what we actually feel and believe. When we struggle to put something into words, that struggle is the thinking. It's how ideas get formed, how emotions get named, how confusion becomes clarity. When AI does that for us, we surely have the text in front of our eyes, but we have skipped the process that would have changed us. The output exists, but the growth doesn't happen.

The same is true for problem-solving. Working through something difficult, sitting with the unknown, trying approaches that don't work, and gradually finding our way towards an answer, is deeply uncomfortable. It is also how confidence, resilience, and genuine competence are built. When AI removes that discomfort by solving things on our behalf, it also removes the experience of discovering that we can figure things out. Over time, that discovery stops happening, and we begin, quietly, to trust ourselves less.

Creative thinking is perhaps the deepest loss of all. 

Creativity is how we make sense of the world, how we express who we are, and how we contribute something that could only have come from us. When we outsource creativity to AI, asking it to generate our ideas, our images, and our stories, we are saving time, of course, but we are also stepping back from one of the most fundamentally human activities there is. The bad news is that the less we do it, the harder it becomes to imagine that we could.

What makes all this particularly dangerous for our mental health wellbeing is that cognitive engagement, such as thinking, processing, writing, creating, and problem-solving, is not separate from emotional wellbeing. The very opposite. It is deeply bound up with it. A sense of purpose, agency, empowerment, and identity all depend, in part, on the experience of doing things, and when AI absorbs those activities, it makes us less capable. It can make us feel less like ourselves.