The last decade has seen phishing scams increasingly leveraging text and instant messages instead of email as attack vectors. Due to fewer diagnostic cues available in text and instant messages and the use of short URLs, generative artificial intelligence and advanced baiting techniques by scammers, the effectiveness of conventional phishing scam detection approaches based on sender information, domain name, image, content and context is being challenged. This chapter describes a novel approach for identifying Chinese phishing text and instant messages based on the persuasion structure in scam plots and explaining to potential scam victims why the text and instant messages are deemed deceptive. A compact phishing content dataset was constructed to cover three common types of phishing scams encountered in Hong Kong, following which the MacBERT large language model was employed to identify persuasion features. The resulting model yields 92.5% overall accuracy in text/instant message corpus classification and 95.7% accuracy in scam detection. Additional testing demonstrates that the model generalizes well to a global English phishing text message dataset, reflecting the universality of scam detection based on persuasion structure.