With the rapid advance of text-mining technology, an increasing amount of valuable information is being used to address issues related to financial distress prediction. However, the utilization and interpretability of textual data in prediction models pose challenges for its practical application. This study proposes a practical text-processing approach to extract predictive information embedded in financial reports. Specifically, the study segments textual documents into individual words and employs term frequency–inverse document frequency (TF-IDF) as the input for prediction models. Experiments conducted on data from 2000 to 2021 for Chinese listed companies demonstrate that textual information exhibits favorable predictive capabilities for both short-term and long-term financial distress. Further, significance tests reveal that textual information serves as an effective supplement to traditional financial data. When compared with financial indicators alone, the introduction of textual indicators improves prediction outcomes by 1–3 percentage points across various prediction scenarios. The study also examines critical words that play a significant role in financial distress prediction, highlighting their potential as signals for identifying financial health or distress. By offering an alternative, practical method, this study contributes to the application of textual information in financial distress prediction.