Using Reuters data and machine learning, this paper explores the potential role of linguistic features of 972,144 business news headlines in going-concern opinion predictions. This research identifies 26,857 firm-year observations from their headlines from 2004 to 2016. For each observation in the sample, 286 linguistic features are extracted from the news headlines, including the frequency of company mentions, sentiment features (identified by using three advanced natural language processing (NLP) tools: TextBlob, Flair, and VADER), and the frequency of financial topics mentions. The most important linguistic features were selected with the Balanced Random Forest algorithm. Next, the selected news features are combined with audit-related variables, financial variables, and market variables incrementally in logistic regressions to examine the association between those variables and the dependent variable, going-concern opinions. Specifically, to simulate the dynamic process of audit at different stages, seven models are developed. The results show that the linguistic features of business news headlines are significantly associated with the going-concern opinions, especially at earlier stages of the audit process, suggesting business news headlines are important resources of additional information regarding both the audit risk and the company’s business risks in the formation of a going-concern opinion. Press coverage information, and specifically news headlines, influences the auditor’s risk perception of a potential going concern issue, the issuance of which may help decrease future litigation risk.