Explainer
AI and Therapy: How AI Supports Mental Health Care
AI in therapy refers to software tools — typically large language models — used to support mental health care alongside human therapists. Responsible platforms use AI for journaling prompts, mood tracking, psychoeducation, and administrative help (like matching or scheduling), while keeping all clinical decisions with a qualified human clinician. MySafeTherapy follows this model: BACP/UKCP/HCPC registered therapists deliver the therapy, and AI is used only for transparent, guardrailed self-help features.
Key points
- AI is best used to augment therapy, not replace it.
- Suitable uses: journaling, mood tracking, psychoeducation, booking, and reflection prompts.
- Unsuitable uses: diagnosis, crisis response, medication advice, or unsupervised long-term care.
- Reputable AI-therapy tools clearly label AI, avoid clinical claims, and defer to humans in crisis.
- Evidence base is early but growing: research shows benefit for mild-to-moderate distress when paired with human oversight.
How AI is used in mental health today
- Guided journaling (e.g. Navi in MySafeTherapy) — AI prompts encourage reflective writing.
- Mood check-ins and pattern spotting — noticing trends across weeks or months.
- Therapist matching — helping new users find the right clinician quickly.
- Content personalisation — recommending relevant guided journeys or booklets.
- Session admin — summaries, reminders, and structured intake forms.
- Between-session support — micro-exercises like breathing, grounding, or CBT worksheets.
Where AI should NOT be used
- Clinical diagnosis — only qualified clinicians can diagnose.
- Crisis or safeguarding decisions — must route to humans and emergency services.
- Medication or dosage advice — regulated medical territory.
- Replacing the therapeutic relationship — the relationship itself is what heals.
How MySafeTherapy blends AI with human care
Every clinical interaction is with a UK-registered therapist. AI tools like Navi (journaling), Lima (booking guide), and Trigger Translator (thought reframing) are clearly labelled as AI, run on guardrailed models, and never make clinical decisions. Sensitive inputs are scanned for safety concerns and routed to human support when needed.