The study, conducted by Franco et al., aims to explore how natural language processing (NLP) and large language models (LLMs) can analyze autobiographical memory narratives in relation to depression and suicide risk. Researchers collected free-text narratives from 915 participants from Brazil and used the Google Gemini LLM to generate text embeddings, which is a numerical representation of text. These embeddings were analyzed using Independent Component Analysis (ICA) and machine learning models to determine the emotional affect (positive or negative) of each memory. The research found that the emotional affect had a moderate-to-high predictive value on the prevalence of participants’ suicidal ideation, depression diagnosis, and past suicide attempts, with suicidal ideation having the most predictive value at 84.3% accuracy. These predictive values improved when combined with a validated psychometric tool, such as the Three-Step Theory of Suicide referenced in the research. The strongest relationship between the emotional affect and suicide risk factors was with a lack of connectedness, which is a core component of Three-Step Theory. Additionally, NLP allowed detection of certain linguistic markers of mental distress that traditional methods might miss. This approach enhances psychological assessment by identifying latent emotional and cognitive markers in narratives, garnering potential for real-world clinical application in suicide prevention.