The effects of artificial intelligence-based chatbots on mental health: A systematic review and three-level meta-analysis.
Qiyi Wang, Jiale Zhong, Kai Qi et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and three-level random-effects meta-analysis assessed AI-based chatbot interventions for mental health across 16 eligible studies, incorporating 67 effect sizes. The pooled estimate showed a statistically significant small-to-moderate improvement in mental health outcomes (Hedges' g = 0.47, 95% CI [0.18, 0.77]); leave-one-out analyses found the estimate stable. Improvements were reported across depression, anxiety, loneliness, affect, and sleep, with no significant moderators identified. However, most studies had some risk of bias and the overall certainty of evidence was low. The findings reflect intervention packages, not the isolated effect of AI technology.
Key findings
- Across 16 studies and 67 effect sizes, chatbot interventions were associated with improved mental health outcomes (Hedges' g = 0.47, 95% CI [0.18, 0.77]).
- Improvements were observed across depression, anxiety, loneliness, affect, and sleep; no significant moderators were identified.
- Leave-one-out sensitivity analyses indicated that the pooled effect estimate remained stable.
Why it matters
The pooled findings suggest that chatbot-based intervention packages may offer modest benefits across several mental health domains. This is relevant amid constrained access to care, but low evidence certainty and risk of bias mean the results should be interpreted cautiously. They do not establish that AI itself, rather than other components of the interventions, produced the observed improvements.
What to keep in mind
Most included studies had some risk of bias, and overall evidence certainty was low. Heterogeneous intervention types also limit attribution of effects to AI technology itself.