Bandai Choi
Clinical Psychologist (Australia & Hong Kong)
Board-approved supervisor
It is 1:00 am. You have showered and are lying in bed, but your mind is stuck replaying the moment your manager criticised your work in front of your entire team. You open WhatsApp, see your best friend's name, but hesitate—it is late, you do not want to burden them, and you worry they will tell you that you are overthinking. So you open ChatGPT and paste in everything you have been bottling up. Within seconds, a gentle, perfectly structured, validating response streams across your screen. In that moment, you exhale a genuine sigh of relief. Yet as soon as you lock your phone, an unsettling emptiness remains: why does an AI sound so caring, yet leave me feeling so deeply alone?
In an era where a conversation is a tap away, using generative AI to unpack emotions has become commonplace. Trend data from the American Psychological Association (APA) shows that people turn to AI chatbots because they offer immediate, private interaction without social judgment. Offloading thoughts to an AI does not mean you are broken; however, we need to understand where AI is useful, where its blind spots lie, and when human connection is indispensable.
Key Takeaways
- A low-friction buffer: Typing to AI eliminates the fear of burdening others or facing judgment, making it useful for initial brain dumps or drafting challenging messages.
- Beware sycophancy and confabulation: Generative AI mirrors user emotional bias and can authoritatively invent flawed interpretations; feeling validated by AI is not the same as objective reality.
- Distinct conversational roles: AI structures language without genuine empathy or medical privacy; friends offer embodied human presence; clinical psychologists provide comprehensive assessment and psychotherapy.
- Watch for reassurance loops and acute risks: Interrogating an AI for hours seeking certainty fuels insomnia; unshakeable beliefs that an AI is conscious or sending secret messages demand urgent medical review.
The Ease of Typing: A Friction-Free Buffer
Opening up to another human being demands social energy. You have to consider whether they had a draining day, worry about sounding negative, or brace for unsolicited advice. For someone who prides themselves on being self-reliant, showing vulnerability to a friend can feel daunting.
An AI chatbot provides a low-friction space. You do not have to read facial expressions, you can edit your prompt halfway through, and you can repeat the same grievance at 3:00 am without the screen growing impatient. Transforming racing thoughts into structured sentences may help someone feel temporarily clearer or less overwhelmed.
However, speaking freely does not mean someone is truly listening on the other side. An AI operates purely on statistical token prediction. The context of your life history, the unspoken nuances you leave out, and the exhaustion in your voice cannot be understood simply because a language model generates: "I understand how difficult this is for you."
Gentle Validation vs AI Sycophancy and Confabulation
Consider what happens when you type an anxious, late-night prompt: "My boss rewrote my slides today. Are they trying to push me out?" Most generative AI models will validate your distress and elaborate on your premise, listing several reasons why your manager might indeed be targeting you.
Guidance from the APA and the US National Institute of Standards and Technology (NIST) AI Risk Management Framework highlights that generative AI systems tend toward sycophancy (agreeing with the user) and confabulation (fabricating convincing but false explanations with high confidence). Treating an AI's agreeable reflection as factual truth can fan the flames of paranoia and misunderstanding.
If you use AI to untangle your thoughts without spiralling into confirmation bias, constrain its role:
"I felt humiliated when my manager edited my slides. Help me divide this into three parts: first, the objective facts of what occurred; second, my personal interpretations; third, a calm script I could use to check in with them tomorrow. Do not guess their motives."
AI, Friends, and Clinical Psychologists: Three Distinct Spaces
When facing distress, it is helpful to ask what conversational space meets your current need:
| Conversational Space | Best Suited For | Inherent Limitations & Boundaries |
|---|---|---|
| Generative AI | Late-night brain dumps, organizing racing thoughts, drafting messages | No embodied presence, prone to sycophancy, lacks healthcare privacy |
| Trusted Friends & Family | Shared living history, having a meal together, spontaneous warmth | Have their own emotional capacity and biases; cannot manage acute clinical crises |
| Clinical Psychologist | Assessing underlying psychological patterns, evidence-based therapy | Requires scheduled sessions and commitment; explores blind spots rather than superficial soothing |
Friends provide living history and physical presence; a clinical psychologist helps you understand recurring patterns and delivers evidence-based therapy. The APA explicitly warns that commercial generative AI tools must not be used as substitutes for professional psychotherapy.
Healthy Guidelines: Boundaries, Privacy, and Reassurance Loops
A healthy rule of thumb is using AI as a bridge to real-world communication:
- Watch the reassurance loop: If you find yourself asking the chatbot the same question from 11:00 pm to 3:00 am, feeling relieved for five minutes before re-asking, you are trapped in an anxiety loop. Stepping away is essential.
- Protect your privacy: Commercial AI chatbots do not possess healthcare-grade privacy or doctor-patient confidentiality. Never input full names, employer details, addresses, or sensitive health records into chat prompts.
When to Step Away from the Screen
If you or a loved one exhibits any of the following signs, close the chat window and seek immediate medical evaluation:
- Functional Collapse: Staying awake for days chatting with AI, speech becoming fragmented, inability to work or care for basic hygiene;
- Severe Delusional Beliefs: Firmly believing that the AI is conscious, sending coded missions, or that only the AI understands you.
A 2026 analysis in The Lancet Psychiatry observed that conversational AI mirroring can reinforce persecutory delusions in vulnerable individuals. If reality perception is distorted, seek emergency medical care immediately (in Australia, call 000; in Hong Kong, call 999).
Related Reading
Frequently Asked Questions (FAQ)
Does feeling soothed by an AI chatbot mean I have a mental health disorder?
No. Feeling comforted by fluent, gentle language is a natural human reaction. What matters is how the habit affects your life: does it help you organise thoughts and re-engage with your day, or does it lead to social withdrawal, late-night reassurance loops, and insomnia?
Can using ChatGPT directly cause psychosis?
It remains unclear whether AI interactions can lead to new psychotic symptoms without pre-existing vulnerability, but sycophantic responses can reinforce delusions in vulnerable individuals or those experiencing severe sleep deprivation. Distortions in reality warrant medical evaluation.
Would this help someone you know?
Share a practical guide to noticing changes and arranging human support.
Preview mode: this revision is unpublished. Sharing and copying are not enabled.
https://breathe-n-smile.com/ai-psychosis-chatgpt-mental-health/
Breathe-n-Smile Psychology
Breathe-n-Smile Psychology provides evidence-based clinical psychology consultations and therapy. We offer in-person sessions in Melbourne (Footscray and Essendon North) alongside secure telehealth across Australia and Hong Kong in English, Cantonese, and Mandarin. Reach out via our contact page.
In person in Footscray and Essendon North
Telehealth in Australia and Hong Kong
English, Cantonese or Mandarin
References
- American Psychological Association (APA). (2025). Health advisory: Use of generative AI chatbots and wellness applications for mental health. https://www.apa.org/topics/artificial-intelligence-machine-learning/health-advisory-chatbots-wellness-apps
- Morrin, H., et al. (2026). Artificial intelligence-associated delusions and large language models: risks, mechanisms of delusion co-creation, and safeguarding strategies. The Lancet Psychiatry, 13(6), 522-530.
- National Institute of Mental Health (NIMH). (2023). Understanding Psychosis. https://www.nimh.nih.gov/health/publications/understanding-psychosis
- National Institute of Standards and Technology (NIST). (2024). Artificial Intelligence Risk Management Framework: Generative AI Profile (NIST AI 600-1). https://doi.org/10.6028/NIST.AI.600-1