Evaluative Meaning in AI-Generated Opinion Texts about Indonesia's DPR

Abstract

Artificial intelligence has become increasingly capable of producing journalism texts, raising questions about how it constructs evaluative meaning. This study explores how AI systems express attitudes toward the Indonesian House of Representatives in opinion pieces written in response to recent socio-political events. Using the Attitude subsystem of Martin and White’s Appraisal Theory, the research focuses on identifying Affect, Judgment, and Appreciation in AI-generated texts. The data were collected from three popular AI platforms, namely ChatGPT, Gemini, and Perplexity, through structured prompts designed to elicit critical opinions. The findings reveal that while AI can articulate evaluations coherently, its expressions are dominated by judgment and appreciation rather than affect, showing limited emotional engagement. This suggests that AI primarily reproduces socially acceptable evaluative patterns rather than genuine feelings.