A systematic review and meta-analysis published in the Journal of Medical Internet Research in December 2025, covering 14 randomised controlled trials and over 6,300 participants, found that generative AI chatbots are effective therapeutic tools for reducing negative mental health issues such as depression and anxiety (effect size = 0.30, P = .047).
Key findings
| Metric | Value |
|---|---|
| Randomised controlled trials | 14 |
| Total participants | 6,314 |
| Effect size (Hedges' g) | 0.30 |
| Statistical significance | P = .047 |
| Studies in narrative synthesis | 26 |
What types of interventions worked best
The review identified several factors associated with stronger outcomes:
- Empathetic, non-judgmental conversation – chatbots that provided supportive dialogue showed more consistent benefits
- Structured therapeutic approaches – tools incorporating elements of CBT and other evidence-based frameworks
- Regular engagement – consistent interaction patterns correlated with better results
Limitations the researchers noted
The authors highlighted that most interventions took place in non-WEIRD (Western, Educated, Industrialised, Rich, Democratic) countries, and there is a lack of studies focusing on young children and older adults. They also emphasised that generative AI chatbots' risks cannot be ignored, calling for further research into ethical implications.
What this means for Companion
Companion is designed around core elements that this research supports: a conversational model tuned for empathy and non-judgment. Combined with automatic diary generation and voice support, it creates a space where users feel genuinely heard. As the evidence base for AI-assisted mental health tools continues to grow, Companion aims to be at the forefront of responsible, evidence-informed support.
Source: Zhang, Q., Zhang, R., Xiong, Y., Sui, Y., Tong, C., & Lin, F.-H. (2025). "Generative AI Mental Health Chatbots as Therapeutic Tools: Systematic Review and Meta-Analysis of Their Role in Reducing Mental Health Issues." Journal of Medical Internet Research, 27, e78238.