Decoding antisemitism online: linguistic and multimodal challenges in the age of AI

Introduction This article investigates the linguistic and computational challenges of detecting antisemitism in digital communication, integrating discourse-analytical and artificial intelligence (AI) perspectives. It conceptualizes antisemitic discourse as a continuum ranging from explicit incitement to implicit, coded expressions whose interpretation depends on contextual, cultural, and pragmatic knowledge. Methods The study draws on empirical case studies from the Decoding Antisemitism project, analyzing YouTube reactions to two events: the Hamas terror attack of 7 October 2023 and the antisemitic double murder in Washington, D.C., in May 2025. Qualitative discourse analysis is combined with computational considerations related to annotation practices and model design for automated detection. Results The analysis shows that antisemitic discourse has become normalized in mainstream digital spaces. Reactions to 7 October were characterized by open glorification of violence, whereas responses to the Washington case centered on denial, irony, and the inversion of victimhood. Together, these cases illustrate both the normalization and diversification of antisemitic communication online. Discussion Building on these findings, the article discusses methodological and computational implications for antisemitism detection. It highlights challenges such as semantic ambiguity, pragmatic drift, multimodal signaling, and data scarcity, and evaluates emerging computational approaches, including transformer-based fine-tuning, retrieval-augmented systems, and context-engineered large language models (LLMs). The study concludes that effectively confronting digital antisemitism requires sustained collaboration between linguists, data scientists, and policymakers to develop context-sensitive, transparent, and ethically grounded AI systems capable of reliable interpretive reasoning.