Abstract: 【Purpose/significance】This study explores AI-driven debate techniques for public opinion intervention, aiming to mitigate the cognitive risks posed by harmful online content and enhance cognitive security in China—a topic currently underexplored. 【Method/process】Using a comprehensive literature review, the paper examines domestic and international research progress from four perspectives: conceptual framework, theoretical foundations, technical bases, and key challenges in debate-based public opinion inter⁃ vention.【Result/conclusion】Existing research emphasizes textual semantic analysis but overlooks non-translatable modalities such as speech, scripts, and rhetorical language, particularly in cross-cultural contexts. Studies on the interplay between online emotions and user cognition are limited, and static data analysis fails to capture dynamic shifts. Debate mining methods rely heavily on probabi⁃ listic models, lacking structural analysis capabilities and unified evaluation standards. Debate generation research remains insufficient in addressing topic complexity and unpredictability, and reinforcement learning is underutilized. Current corpora are scenariospecific, lacking generality, scale, and robust Chinese debate data, with issues such as concept drift. These gaps outline critical direc⁃ tions for future work.【Innovation/limitation】This study provides a systematic review of public opinion intervention via debate tech⁃ nologies and identifies key areas for future exploration, including causal reasoning in autonomous AI debate, knowledge incomplete⁃ ness in deep language understanding, and challenges in style and semantic invariance in natural language modeling.