Artificial Intelligence and Mental Health Discourse: Methodological Reflections for Nursing Practice

Social media offers large-scale access to mental health discourse, enabling population-level insights into distress, support-seeking, and community dynamics relevant to person-centered care. Artificial Intelligence (AI) methods can analyze naturally occurring expressions to complement clinical understanding by identifying patterns across individuals and groups. This study examines interaction-based classification and sentiment analysis approaches—including traditional machine learning, transformer models, and Large Language Models (LLMs) applied to Twitter data from self-disclosing users across multiple mental health conditions. Interaction features showed moderate ability to distinguish mental health -related accounts from controls but limited specificity across conditions. Sentiment patterns were broadly consistent across models, with mostly neutral discourse and higher negative sentiment in some diagnostic communities. However, LLMs demonstrated sensitivity to prompting and occasionally produced inconsistent outputs, raising concerns about reproducibility, interpretability, and ethical use. Overall, AI approaches show promise but remain probabilistic and context dependent. They should complement, not replace, clinical judgment. Transparent methods, attention to bias, and ethical oversight are essential to ensure these tools support equitable, person-centered mental health care.