The Efficacy of Rule-based Versus LLM-based Chatbots in Alleviating Symptoms of Depression and Anxiety: A Systematic Review and Meta-Analysis.

Comprehensive Summary

This article, presented by Du et al., evaluates the differences in effectiveness between LLM-based and rule-based chatbots at alleviating depression and anxiety symptoms through systematic review and meta-analysis. Literature screening was conducted in two stages: initial screening of titles and abstracts and fine screening of full text. During initial screening, AI batch analysis was used in combination with manual review to select literature that would enter the next phase. During the fine screening stage, the authors read and evaluated the full text according to the inclusion and exclusion criteria. 8834 articles published between 2020 and 2025 across 7 databases were initially identified, and 15 were ultimately included in the meta-analysis. The overall meta-analysis indicated that rule-based interventions produced small but statistically significant improvements in depressive symptoms and no significant improvements in anxiety symptoms; LLM-based interventions did not show significant improvement for either group of symptoms. Wide confidence intervals and heterogeneity indicate uncertainty in results. The results also indicated that interventions lasting 4-8 weeks may be more effective. Rule-based chatbots may be limited in more complex and acute anxiety-inducing situations, causing their worse performance for anxiety symptoms. Uncertainty regarding the effectiveness of LLM-based chatbots was mainly due to the small sample size of studies, and larger, more rigorous and consistent studies are needed.

Outcomes and Implications

The prevalence of anxiety and depression is very high and continues to rise, indicating the need for more accessible interventions. AI chatbots are a promising solution due to their affordability, accessibility and ability to help people that cannot access other forms of mental health care. This article suggests that LLM-based and rule-based chatbots have unique strengths that are appropriate in different situations and that hybrid chatbot architectures may be solutions for the shortcomings of both models. However, the study has several limitations, including methodological limitations and the small number of studies included, a high level of heterogeneity among studies, an insufficient exploration of intervention mechanisms and a lack of direct comparison between AI interventions and standard treatments. Further research is recommended to increase the sample size of studies, conduct high-quality trials and explore moderating factors.