In the digital age, the dissemination of Chinese textual information on the Internet significantly influences people’s decisions and judgments. Online reviews, introductions, and comments affect daily choices, especially with the popularization of online shopping, where purchasing decisions often rely on product reviews. However, the authenticity of these reviews affects their quality, leading to potentially inaccurate information for users. This study utilizes Google BERT’s language recognition capabilities to identify the authenticity of product reviews for general consumer goods. It focuses on five types of product reviews from Amazon: household items, electronic products, clothing, toys, and pet supplies. By training the model and combining it with web crawling, the study filters out fake textual information. This filtered information is analyzed using the LDA topic model to explore its structural meaning and validate definitions of fake reviews. The research reveals frequently repeated vocabulary across the five domains. High repetition rates within topic blocks result in a lack of detailed information, complicating topic identification. Sentiment analysis shows a positive bias in reviews, significantly higher than neutral. These findings confirm characteristics of fake reviews, such as frequently repeated vocabulary, lack of detailed information, and the use of extreme words. Thus, BERT’s large language model proves highly feasible for identifying fake reviews.