Societal threats such as pandemics, crises, and political instability are believed to influence public trust, yet their long-term linguistic effects remain underexplored. This study examines whether the prevalence of threat-related language can predict shifts in trust-related discourse over extended time periods. Using word2vec trained on Google News (English) and Weibo corpora (Chinese), we identified high-similarity terms associated with “threat” and “trust” in both languages. Annual frequencies of these terms were extracted from the Google Books Ngram Viewer (1945–2019 for English; 1970–2019 for Chinese). Linear regression and granger causality analyses revealed that increases in threat-related language could predict decreases in trust-related terms, with 1–2-year lags in English and 3–5-year lags in Chinese. These findings suggest that threat discourse can serve as a predictive signal of conceptual trust erosion, offering a cross-linguistic framework for tracking socio-psychological change via large-scale textual data.