How AI’s reporting Differs: Evidence of Themes, Emotions, and Representation of Social Actors

The rapid emergence of generative AI models in the media sector necessitates a critical examination of the reporting they produce. In this context, we focus on the refugee crisis arising from the Russia–Ukraine conflict, a highly contentious global issue, as the subject of analysis. Employing an interdisciplinary methodology, this study integrates corpus tools, LDA modeling, emotional analysis, and critical discourse analysis to systematically compare the reporting patterns of three generative AI models (GPT-4, Claude 3.5 and Gemini1.5) with three mainstream media agencies (CBS, CNN and Fox). Analyzing 150 media reports and 150 AI-generated articles, our quantitative and qualitative findings reveal distinct differences in reporting on Ukrainian refugees. AI models tend to avoid political reporting and adopt a more “humanized” tone, with lower scores for fear emotions and higher anticipation scores compared to traditional media. In terms of the representation of social actors, AI-generated content demonstrates a greater reliance on categorization and assimilation. These findings provide critical insights into the unique characteristics of AI-generated reporting and their potential implications for media discourse.