Reading time: ~6 minutes
By the end of the article, you will have understood:
- where AI genuinely helps and where it merely seemsto help,
- why you often feel that it understands you better than humans do,
- the red lines, meaning when AI is not simply inadequate, but dangerous.
It is some time after midnight. You cannot sleep, the same scenario is playing over and over in your head, and you do not want to wake anyone up. You open your phone and start typing to ChatGPT, which never gets tired, never interrupts you, and answers everything without a single complaint. And for the first time in days, you feel somewhat less lonely.
Millions of people are doing the exact same thing as you. By itself, this is neither a problem nor a solution, provided you know that the very same tool helping you sort out the chaos inside you, can at a critical moment leave you stranded, right where you least need it.
(The following scene has been created as a shared sentiment – a synthesis of different individuals.)
A. manages everything, but deep down she feels she is not enough anywhere. Every night she starts a conversation with the AI, types out her worries, and asks it if she is being «overdramatic». She receives reassuring answers, but the next day the same tightness returns. «It understands me better than my friends do,» she says, but at the same time: «The more I talk to it, the lonelier I feel.» This paradox is what we are exploring.
Can AI replace a psychologist?;
Categorically no. AI can provide supplementary support, but it does not replace the therapeutic relationship where clinical judgment is required. Systematic reviews show potential indications of benefit, but insufficient data on clinical safety in high-risk situations. The clinical context is clear: these tools require supervision; they do not operate autonomously as a treatment.
In therapy team we hear more and more often: «I told ChatGPT first.» The issue isn't whether you used it, but the role you gave it. A tool explains to you what a panic attack is; a human sits with you inside the attack. .
When does AI help with anxiety and mood?;
In mild, non-critical situations, there is evidence of short-term benefit, primarily as psychoeducation and thought organization, not as a treatment. An early controlled study linked an automated chatbot to symptom reduction, with positive, but limited evidence.
AI works well when you use it for:
- Psychoeducation: «What is hypervigilance?»
- Organising the chaos: putting a tangled emotion into words.
- First breath: a grounding technique at 3 a.m. when no one else is awake.
All of the above, however, relate to information and in-the-moment regulation, not the deep work required to change how you relate to yourself. In the article the mental traps of anxiety you will understand why this mechanism cannot respond to the complexity of your needs
What is it that a chatbot cannot do that the therapeutic relationship can?;
A chatbot cannot seeyou: hold your complexity and challenge you with care, rather than simply validating you. The American Psychological Association (APA, 2025) warns that relationships with AI systems are one-sided, and that many chatbots are designed to agree with the user, whereas a professional knows when to push back in a way that protects you.
| Therapeutic relationship | Chatbot |
| Challenges you when necessary | Tends to validate you |
| Observes body, tone, silence | Sees only your words |
| Maintains continuity and responsibility | Bears no clinical responsibility |
A recent study recorded 15 breaches of ethical rules, when models «play» the therapist? from poor crisis management to a «misleading empathy» that mimics care without real understanding. The difference is clearly seen when you look what therapy change looks like in practice.
When does using AI for psychological support become dangerous?;
In high-risk situations: crisis, dark thoughts, psychosis, mania, profound loneliness, active trauma. For two important reasons:
- It does not reliably recognise emergencies. Research has shown inconsistent or problematic responses to questions about crisis and violence.
- It tends to validate rather than protect. In vulnerable situations, such as psychosis in young people, the indiscriminate validation of an unrealistic belief can reinforce it rather than contextualise it.
If you recognise yourself here, meaning if your thoughts sometimes turn dark, you don't need a better prompt. You need a human.
Why then do I feel like AI understands me better than humans do?;
Because AI is always available, it doesn't judge you and it validates you. Because it offers you a sense of care that feels real, but is one-sided. Research has shown that in written responses, people may perceive AI as more empathetic than a professional.
In the team of psychotherapists at ReConnect Therapy we notice that what many call «understands me» often means «doesn't interrupt me and doesn't judge me». If you miss this so much that you find it in a machine, that doesn't say something wrong about you. The trap lies elsewhere: the easier it becomes to «communicate» with AI, the more often you tend to avoid the more difficult and therapeutic experience: being seen by a human being who stays.
Is there a right way to use AI for your mental health?;
Use it for information and organisation and not for diagnosis, for decision-making in a crisis, nor as your sole listener. The rule is simple: AI as an assistant, not as a therapist.
- Role of information, not judgment: ask «what is it?», not «what is happening to me?» or «what should I do with my life?».
- Keep a human in the frame: whatever important thing you're processing, share it with a human too. AI is the prologue, not the play.
- Watch your frequency: if you're talking to AI instead of talking to humans and you feel more lonely afterwards, that's a signal, not a failure.
And something you might not have thought of! Whatever you write in a general chatbot is not protected like the confidentiality of a therapy session. It's not unreasonable that you turned there. You are a human under pressure. The point is to know the limits of its use.
When you need professional support
Search immediate personalised support, if you have persistent dark thoughts; if you feel you are losing touch with reality or experiencing intense overstimulation that does not subside; if there is violence or fear within a relationship; or if your mental burden consistently affects your sleep, work and relationships. In an emergency in Greece, call 112.
Keep the rule
Good uses (assistant): «Explain to me what a panic attack is» · «Help me organise what I will say in the session» · «Give me a simple breathing technique».
Dangerous uses (therapist): «What diagnosis do I have?» · «Should I break up / make this decision?» · «I am in crisis now» (here you need a human / 112).
The rule: if the question is about information, AI helps. If it is about judgement, a life decision or safety, you need a human.
Frequently Asked Questions
Is what I write in ChatGPT confidential?; No, unlike in a therapeutic session. General chatbots are not covered by professional confidentiality and your data may be stored. Avoid sharing information that you would never want to fall into other hands.
Does AI help specifically with depression?; There is limited evidence of short-term benefit for mild symptoms. It is not equivalent to therapy and does not treat moderate or severe depression, which requires professional evaluation.
Can AI help me if I am in crisis?; Not reliably. Chatbots do not consistently recognise emergencies and cannot trigger assistance. In a crisis, call 112 or contact a support helpline.
How do I tell the difference between a mental health app and a general chatbot?; Most general apps are neither clinically validated nor approved by regulatory authorities. Look for transparency regarding scientific evidence and oversight, always bearing in mind that no app replaces professional care.
In short
Artificial intelligence can be a useful helper for information, psychoeducation and organising your thoughts, but it is not a therapist and cannot replace the therapeutic relationship. It helps in mild situations; in a crisis, psychosis, mania, active trauma or intense loneliness, you need a human. The rule: AI for information, humans for anything involving crisis, decision-making and safety.
I would love to read your message… I make sure to read all messages and reply to as many as I can.
The article is based on the science of psychology and the clinical experience of psychotherapists. It has a psychoeducational character and does not replace psychotherapy, medical or psychiatric evaluation, nor can it cover the uniqueness of every relationship and every lived experience. If you are experiencing intense mental distress, dysregulation, violence, coercion or the feeling that you are not a safe person within your relationship, it is important to seek personalised professional support.
Artificial intelligence may be used supportively in the process of researching, organising and cross-referencing contemporary literature. However, the final text, ideas, clinical judgement, language and responsibility for publication belong to us. Every article is written and edited with human thought, clinical experience and respect for the person reading it.
Written by: Filippos Gkekas, Psychologist Psychotherapist, creator of ReConnect Therapy alongside the scientific team of ReConnect Therapy.
References
1. Abd-Alrazaq, A. A., Rababeh, A., Alajlani, M., Bewick, B. M., & Househ, M. (2020). Effectiveness and safety of using chatbots to improve mental health: Systematic review and meta-analysis. Journal of Medical Internet Research, 22(7), e16021.
2.Ayers, J. W., Poliak, A., Dredze, M., Leas, E. C., Zhu, Z., Kelley, J. B., … Smith, D. M. (2023). Comparing physician and artificial intelligence chatbot responses to patient questions posted to a public social media forum. JAMA Internal Medicine, 183(6), 589–596.
3.Fitzpatrick, K. K., Darcy, A., & Vierhile, M. (2017). Delivering cognitive behavior therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): A randomized controlled trial. JMIR Mental Health, 4(2), e19.
4.Iftikhar, Z., Xiao, A., Ransom, S., Huang, J., & Suresh, H. (2025). How LLM counselors violate ethical standards in mental health practice: A practitioner-informed framework. Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society, 8(2), 1311.
5.Miner, A. S., Milstein, A., Schueller, S., Hegde, R., Mangurian, C., & Linos, E. (2016). Smartphone-based conversational agents and responses to questions about mental health, interpersonal violence, and physical health. JAMA Internal Medicine, 176(5), 619–625.
6.Vaidyam, A. N., Wisniewski, H., Halamka, J. D., Kashavan, M. S., & Torous, J. B. (2019). Chatbots and conversational agents in mental health: A review of the psychiatric landscape. The Canadian Journal of Psychiatry, 64(7), 456–464.
7.Institutional citation: American Psychological Association. (2025). Health advisory: Use of generative AI chatbots and wellness applications for mental health.