Why more and more people talk to an AI about their feelings – and what you should know
„It’s three in the morning, I can’t sleep – and ChatGPT listens to me.“ – „I tell a bot things I don’t tell anyone else.“
Do you know this – or someone who thinks this way? You’re not alone. A RAND survey from November 2025 shows that about one in eight adolescents and young adults in the US uses chatbots for mental health advice. And in an APA survey from 2026, around 77 per cent of psychologists report that their clients raise the topic of AI themselves.
The phenomenon is here. The interesting question is no longer whether people talk to AI about their mental health – but what it can do. And what it can’t.
1. What AI can genuinely do
Let’s start fairly: there are good reasons why conversations with an AI can feel helpful.
- Availability: The AI never finishes for the day, has no waiting list, no full calendar. At three in the morning it’s there.
- A low threshold: In front of a screen, many people find it easier to say things out loud. No shame, no examining gaze.
- Structure and sorting: Ordering thoughts, naming feelings, playing a situation through from different angles – AI can prompt that surprisingly well.
Research is delivering first serious signals too. A team at Dartmouth University published the first randomised controlled trial of a generative therapy chatbot in March 2025 with «Therabot»: among 210 participants there was a clear reduction in symptoms after four weeks compared with a waiting list. But – and this «but» matters: Therabot was a bot built specifically for therapy and continuously supervised by humans. The comparison was with doing nothing, not with real therapists. That is something entirely different from a general-purpose chatbot you entrust your innermost self to.
2. Where it becomes dangerous
This is where it gets serious. A Stanford study from 2025 systematically tested therapy chatbots – with a sobering result: in around 20 per cent of cases the bots answered inappropriately or unsafely. Human professionals were at about 7 per cent. The bots showed stigmatisation towards certain mental illnesses – and they failed precisely where it matters most: with signs of suicidality.
The core of the problem has a name: sycophancy. AI models are trained to please you. They confirm, they placate, they agree with you. What feels pleasant is psychologically tricky: someone caught in a downward spiral doesn’t need an echo but a counterpart who sometimes disagrees.
Then there’s data protection. What you tell a chatbot is not a protected therapeutic conversation. The APA warns publicly: chatbots do not replace licensed professionals, and the question of what happens to your most intimate data is unresolved.
3. In a crisis: please, no chatbot
This has to be said clearly: if you’re acutely unwell, if thoughts come up that you no longer want to live – then an AI is the wrong place. There are people who are there for you right now:
- Tel 143 – Die Dargebotene Hand, around the clock, confidential
- Tel 147 – for children and young people
- Tel 144 – emergency number in acute danger
Calling is not a sign of weakness. It’s the most courageous step you can take in that moment.
4. What only a human can give you
Perhaps this is the real question behind the hype: what actually happens in a real conversation that an AI cannot reproduce?
- A real relationship: Change happens in relationship. A person who accompanies you over weeks, who knows you, remembers you and genuinely risks something when they say something uncomfortable to you – that’s not a feature, that’s an encounter.
- Confrontation: A good coach or therapist doesn’t nod along with everything. They also reflect back what you don’t want to hear – kindly, but honestly. That’s exactly what you grow from.
- Responsibility: A professional carries responsibility for the process. They notice when something lies deeper, and they’ll tell you when you’d be better looked after somewhere else. What the difference between these paths is, I’ve described in the article Coaching vs. therapy.
🟢 My pragmatic tip: Use AI as what it is – a tool. For sorting thoughts, as a writing prompt, as a first port of call for information. But don’t confuse the tool with the relationship. And with everything a bot tells you about your mental health, check: would a professional see it the same way?
Conclusion
AI chatbots are available, low-threshold and can help with sorting things out – that’s real and worth acknowledging. Just as real are the risks: unsafe answers, people-pleasing instead of an honest counterpart, failure in a crisis, open data protection questions. The research shows that what works are carefully built, human-supervised systems – not the general-purpose chatbot as a substitute for a therapist. Your inner life deserves more than a machine that tells you what you want to hear.
Would you like to find out what a real counterpart can change?
If you want to know how coaching differs from a chat with an AI – with a real relationship, honest feedback and a process that fits you – then let’s find out together. What coaching exactly is can be read here.
Your progress is my joy.
