Linguistic dynamics of online scam conversations: a multi-stage analysis based on the COLD framework

The global rise of online scams poses escalating social risks, yet the underlying linguistic dynamics remain underexplored. Grounded in the Contextual Organization of Language in Deception (COLD) framework, this study applies a multi-stage analysis of 276,762 messages from 49 authentic scam cases, revealing how language is strategically adapted across five sequential phases—Trust Cultivation, Risk Fabrication, Persuasive Pitching, Compliance Induction, and Repeated Exploitation. Results indicate that linguistic patterns in online scams are shaped by both communicative roles and interactional stages. Scammers consistently employed more cognitive mechanism terms, discrepancy markers, social process words, and affiliation language. They deployed persuasive tactics by strategically increasing achievement-, power-, and money-related terms to project control and success, while suppressing risk-related language. Stage-based analyses indicated that scammers maintained stable linguistic patterns with strategic peaks aligned with manipulation goals, whereas victims exhibited greater emotional and cognitive variability. Victims’ negative emotions and cognitive shifts peaked at critical moments, reflecting resistance and internal conflict. Notably, victims expressed significantly more positive emotion words than scammers, a counterintuitive pattern suggestive of affect-regulation strategies. The findings substantiate a dynamic, multi-stage description of language-based manipulation in online scams and advance the contextual application of the COLD framework. The study also offers practical implications for the development of psycholinguistic risk indicators, automated detection tools, and public education strategies aimed at enhancing resilience against online scams.