Blog
Tips and insights on mental health and emotional well-being
Burnout: the 3 ICD-11 dimensions and what helps
What burnout is per the ICD-11: exhaustion, cynicism and reduced efficacy. How to recognize each dimension, what helps and when to see a professional.
How to cope with stress: techniques by timeframe
Evidence-based self-help techniques for stress — from grounding in a minute to habits over a week — and when stress is chronic and it's time to see a professional.
AI vs. Therapist: A Role-by-Role Map of What 2024–2025 Evidence Actually Shows
A therapist plays four roles. AI in 2024–2025 reaches near-human performance on two (technique delivery, parts of alliance), is triage-only on a third (clinical judgment), and cannot own the fourth (case-level diagnosis).
Five CBT Chatbots, Five Design Choices: How 2024–2025 Studies Map the Field
Five clinically evaluated CBT chatbots in 2024–2025 — SuDoSys (WHO PM+), a cognitive-restructuring system, Socrates 2.0, a BA chatbot, and a GPT-4 PST chatbot — each implements a different rail against the central failure mode: directiveness drift.
CBT-I in an AI Chatbot for Insomnia: A Meta-Analysis of 29 RCTs and an Eight-LLM Experiment
Digital CBT-I reduces insomnia severity at SMD = −0.71 across 29 RCTs (n = 9,475). Bao et al. (2025) tested 8 LLMs on a CBT-I task and showed how to make a chatbot safe.
Therapeutic Alliance With an AI Therapist: 527-User Study
WAI-SR score 3.76/5 with an AI chatbot — near in-person therapy levels. Who bonds with AI, where the alliance breaks, and what actually builds trust.
How AI Detects Suicide Risk in Text — and Where the Method's Limits Lie
NLP models predict suicide risk from linguistic markers more accurately than traditional questionnaires. We examine what AI can do — and where its competence ends.
MIND-SAFE: The Safety Standard for AI Assistants in Clinics and Private Practice
MIND-SAFE (Boit & Patil, 2025) is a three-pillar framework clinics and therapists should use to vet and deploy AI assistants. Breakdown with a checklist.
Prompt Engineering for AI Therapists: Why Off-the-Shelf LLMs Aren't Enough
Only 43% of AI mental health systems include safety measures. The Boit & Patil (2025) framework proposes a three-tier prompt architecture for safe AI therapy.
Do Rule-Based Chatbots Beat LLMs for Depression? A 2025 Meta-Analysis
Meta-analysis by Du et al. (2025): rule-based chatbots moderately reduce depression, while LLM chatbots do not. We break down the paradox and what's behind it.
AI Diagnosis with DSM-5: Transparency Instead of a Black Box
DSM5AgentFlow is a multi-agent system of three AIs that screens mental health conditions using DSM-5 criteria with full justification for every conclusion. Accuracy up to 94%.
Guardrails for AI Therapists: How to Protect Users from Harm
The EmoAgent study (Princeton, 2025) found that 34% of chatbot interactions worsen mental health. The EmoGuard system reduced clinically significant harm to 0%.
Do AI Therapists Work? Meta-Analysis of 35 Studies
35 studies, 17,000+ participants: AI chatbots measurably reduce depression and distress. The effect sizes, the limits, and who benefits most.
Why a Multi-Agent AI Therapist Is 42% More Effective Than a Regular Chatbot
Removing a single agent from a multi-agent AI system reduces therapeutic effectiveness by 42%. We break down why architecture matters more than model size.
Be Your Own Therapist: How AI Teaches Self-Compassion Through Inner Dialogue
Research shows that comforting a virtual 'inner self' improves emotional well-being 13% more effectively than traditional counseling. Here's how the mechanism works.
A Therapist in Your Pocket: Why Running an AI Therapist Directly on Your Phone Matters
An AI therapist that works offline and never sends data to servers. How on-device models protect the most sensitive data you produce — and what regulators require.
How a Small AI Model Outperformed Giants in Psychotherapy
A 500M-parameter model beat GPT-4.1 in therapeutic dialogues. Why size isn't everything in AI-powered mental health support.
AI Ethics in Psychotherapy: Who's Responsible When an Algorithm Harms a Patient?
Who's liable when an AI therapist causes harm? We examine real incidents, FDA regulation, the EU AI Act, and what every user should know.
AI Therapist Reduced Depression Symptoms by 51%: What the First Clinical Trial Found
The first clinical trial of an AI therapist showed a 51% reduction in depression. We break down the research: what's proven and what's still an open question.
ChatGPT as Therapist: Opportunities and Risks of Large Language Models
Millions already discuss their problems with ChatGPT. What does science say about where general LLMs help in therapy — and where they fail dangerously?
Computational Models of Mental Disorders: How Computational Psychiatry Works
Computational psychiatry describes schizophrenia, depression, and anxiety as specific failures in the brain's prediction machinery. Here's how the models work.
Digital Phenotyping: How Your Phone and Your Words Reveal Depression
From GPS traces to voice biomarkers, AI can measure mental state passively. Here's what phones and language actually reveal about depression — and the limits.
Just-in-Time Interventions: How AI Is Learning to Help During a Crisis
What if an app could support you at the exact moment you need it — automatically? We explain the JITAI concept and how AI is learning to respond to stress in real time.