Toward clinical integration of generative AI in mental health: personalization, multimodality and inter-entity experience

Abstract

The integration of conversational agents (CAs) into mental health care presents a promising yet complex frontier. These systems have demonstrated potential in expanding access, enhancing user engagement and providing scalable support, particularly in contexts where human clinicians are unavailable. Despite these advantages, CAs face two fundamental limitations: their quality of interaction and the ability to facilitate self-reflection. This perspective article provides a theoretical framework with clinical relevance, identifying the critical challenges that must be addressed to transition Generative Artificial Intelligence (GenAI) from a conversational tool to an adjunctive psychological support. Firstly, personalization is essential. Advances in retrieval-augmented generation (RAG), fine-tuning, and emotional reasoning are necessary to enable context-aware and ethically grounded responses tailored to individual users. In addition, multimodal interaction (particularly through improvements in speech synthesis, prosody, and expressive dialogue) can help bridge the gap between human and AI communication, fostering greater emotional resonance and natural flow. Lastly, immersive environments, including embodied CA and virtual reality settings, may amplify presence, potentially engaging neural and psychological mechanisms typically associated with human-to-human interaction. These innovations must be accompanied by a strong ethical and regulatory foundation. Systems must ensure transparency, informed consent, and compliance with data privacy standards such as GDPR and HIPAA. Crucially, AI should not be viewed as a replacement for psychologists, but as an adaptive and supportive layer within a broader care ecosystem. By aligning technological capabilities with clinical intent, the future of GenAI in mental health may lie in its ability to complement human expertise and meaningfully extend psychological support.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 2 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.