March 27, 2026 · 7 min read
What AI Can and Cannot Do When You Are Struggling With Your Mental Health
Rochelle Marashi, Therapist
As AI search becomes the first place people turn when they have a question, the mental health space raises important questions about what happens when those questions become deeply personal. Rochelle Marashi is a therapist who has looked at this honestly and without agenda. This is her take.
Many people are now using ChatGPT and similar AI tools to ask questions about their mental health, relationships and emotional wellbeing. That is understandable. AI is available at any hour, costs nothing and does not require you to speak to another person.
But there is an important distinction between what AI can genuinely help with and where it falls short.
What AI Does Well
AI tools have real strengths in this space. They can explain concepts like anxiety, depression, trauma responses and attachment styles clearly and accessibly. They can help you understand what to expect from therapy and how different conditions are typically treated. They can help you articulate feelings and thoughts you are struggling to put into words. And they are available around the clock without requiring you to speak to anyone.
For straightforward informational questions AI can be a genuinely useful starting point.
What AI Cannot Do
The limits matter and they are significant.
AI has no therapeutic relationship with you. The relationship between therapist and client is consistently shown to be the most important factor in therapy outcomes. AI cannot provide that.
AI does not know your history, your circumstances, your relationships or the nuances of your situation. It responds to the words you type, not to you as a person.
AI cannot assess whether you are at risk. It may miss signs of danger, minimise serious concerns or fail to escalate when professional intervention is needed.
AI is not bound by an ethical framework, does not attend supervision and has no professional body overseeing its conduct.
Your conversations with AI may be stored, reviewed or used to train future models. This is fundamentally different from the legal and ethical confidentiality of therapy.
And AI can generate inaccurate, misleading or harmful responses. It may validate unhealthy patterns, provide inappropriate reassurance or miss the severity of a situation.
Where It Gets Risky
Using AI becomes potentially harmful when you are in crisis or experiencing thoughts of self-harm. AI is not equipped to keep you safe. It is also not appropriate when you are dealing with trauma, abuse or coercive control, or when you are using it as a substitute for professional help rather than a supplement to it.
How To Use It Sensibly
If you are using AI to explore questions about your mental health, that shows initiative. Use it for information and education rather than emotional support or decision making. Be aware that responses are generic and do not account for your specific situation. Do not share identifying or highly sensitive information. And if your questions are becoming more serious or distressing, speak to a professional.
If you are in crisis, do not rely on AI. Contact emergency services on 999.
About the Author
Rochelle Marashi is an integrative online psychotherapist offering counselling and psychotherapy to adults navigating life transitions, relationship challenges, stress, trauma and emotional difficulties. She has particular experience supporting clients affected by coercive control, stalking, workplace discrimination and institutional betrayal. Rochelle offers flexible, confidential online sessions and works with clients in five languages: English, French, German, Persian and Spanish. No GP referral is required. rochellemarashi.com
Adam Parker adds a short editorial introduction to all Guest Opinion pieces. Adam Parker is the Founder of AI Search News and Rank4AI, a UK AI search and AI SEO agency focused on improving visibility and recommendation likelihood across ChatGPT, Claude, Google AI Overviews, Gemini, Perplexity and Copilot. rank4ai.co.uk
