A study published in Nature Medicine on March 12, 2026 presents a cognitive layer architecture to support the performance of large language models in psychotherapeutic interactions. This architecture adds specialized psychotherapeutic reasoning capabilities to general chatbots. A real-world study showed that the introduction of this layer improved symptoms of depression and anxiety. Performance was better compared to chatbots alone without a layer. The improvement was also better than that of human therapists. Architecture thus supports more effective psychotherapeutic interactions.