This study systematically reviews the use of natural language processing (NLP) to identify and predict psychosis in individuals at clinical high risk (CHR-P). Researchers gathered and analyzed studies from PubMed, Scopus, and Embase up to May 2025 that applied NLP techniques to speech data from CHR-P individuals, with the goal of determining diagnostic and prognostic accuracy. The final analysis included nine studies, covering 353 CHR-P individuals and 197 controls, using validated diagnostic criteria like the Structured Interview for Psychosis-Risk Syndromes/Scale of Prodromal Symptoms (SIPS/SOPS) and Comprehensive Assessment of At-Risk Mental States (CAARMS). These studies used various NLP models, such as Latent Semantic Analysis (LSA), Word2Vec, Sentence-BERT, and graph-based methods to analyze speech coherence, semantics, and syntax. Diagnostic accuracy ranged from 56% to 95% (AUC 0.86–0.99), and predictive accuracy for psychosis transition ranged from 83% to 100%. Despite promising accuracy levels, the review noted significant methodological diversity, small sample sizes, and limited generalizability, particularly due to reliance on English-trained NLP models and inconsistent validation techniques.