Supportive AI in Mental Health Clinical Text, Screening, Conversational Support, and Documentation in Latin America

Artificial intelligence is being studied in mental health for text analysis, assisted screening, structured conversational interventions, risk prediction, and documentation. The dominant evidence stage remains retrospective development, early evaluation, or short studies in selected populations. Evidence is relatively more developed for natural-language processing, text classification, and structured chatbots targeting depression or anxiety symptoms. Supervised generative documentation is an emerging opportunity whose operational risk may be bounded through human review, not a clinically validated capability. Overall, external validation, prospective clinical utility, impact on processes or outcomes, and routine implementation remain limited or heterogeneous. The evidence does not support independent diagnosis, unsupervised generative therapy, reliable individual suicide prediction, or automated crisis management. Selected studies from Brazil, Argentina, Colombia, and Chile support preliminary transferability, retrospective prediction, feasibility, or bounded evaluation, not regional effectiveness or implementation readiness. Cultural and linguistic bias, privacy, crisis escalation, and human accountability are design requirements. Current defensible uses are limited to reviewable support with bounded risk; applications that may alter diagnosis, treatment, or crisis response require a prior research agenda and clinical and institutional approval. This is a narrative community evidence brief, not a formal scoping review or clinical guideline.