AI-Assisted Multimodal Discourse Analysis Learning Model to Enhance Critical Literacy Skills

Abstract

Background and Objectives: Critical literacy, a key four C's (4C) competencies, is essential for navigating digital complexities. In today’s context, critical literacy goes beyond understanding texts, involving the ability to interpret meanings across modes, uncover underlying ideologies, and critically evaluate the intent and credibility of persuasive or manipulative messages. Yet discourse analysis in Indonesian universities remains confined to linear texts, limiting engagement with multimodal discourse. Emerging artificial intelligence (AI) offers promising opportunities, though its pedagogical applications remain underexplored. AI has the potential to serve as a learning assistant, capable of performing keyword extraction or topic modeling, bias analysis, image–text relation mapping, and sentiment analysis. Despite recognition of multimodal literacy and AI-assisted learning, structured pedagogical models are absent. This study introduces Discourse Interpretation through Systematic Critical Evaluation with Responsive AI-Navigation (DISCERN-AI) model, an AI-integrated model to scaffold reflection, enhance multimodal literacy, and enrich higher education. Methodology: This study employed a research and development (R&D) design involving experts, lecturers, and 120 fourth- and fifth- semester students from four universities in Bali. Data collection focused on validity, practicality, and effectiveness. Model validation was conducted by five experts using Aiken’s V, while practicality and effectiveness were examined though Likert-scale questionnaires and pre–post testing. Data analysis employed a mixed-methods approach, combining ANCOVA, effect size estimation, and thematic analysis to ensure comprehensive triangulation. Results: The DISCERN-AI model comprises six phases, systematically designed as an operational framework for facilitating multimodal analysis in the classroom. Construct validity testing indicated a very high level, V value of 0.88, while content validity analysis yielded a mean coefficient of 0.86, categorized as very valid. Practicality testing by lecturers of the Discourse Analysis course further confirmed the model’s applicability, obtaining a mean score of 3.90, categorized highly practical. Effectiveness was evaluated through effect size analysis using Cohen’s d. Comparison of posttest results between experimental and control groups produced an effect size 1.32, which falls into the large category. Discussions: The DISCERN-AI model brings together critical pedagogy, multimodal discourse analysis, and AI-assisted learning through responsive navigation and multimodal triangulation. Findings from empirical testing show high levels of construct and content validity (V = 0.88 & V = 0.86), strong practicality (M = 3.90), and notable effectiveness (Cohen’s d = 1.32). These results highlight the model’s solid potential to foster critical literacy skills in higher education, aligning with the growing global emphasis on technology supported critical pedagogy. This model guided students to examine textual representations across various modes—verbal, visual, and audiovisual—thereby fostering their interpretive, evaluative, and synthetic skills. Conclusions: The DISCERN-AI model demonstrates strong theoretical validity, practical feasibility, and pedagogical effectiveness in enhacing students’ multimodal critical literacy. Aligned with critical and multimodal discourse analysis frameworks and 21 st-century competencies, the model contributes to theories of multimodal and critical digital literacy. Therefore, DISCERN-AI can be positioned as an innovative pedagogical model that addresses the demands of critical literacy in the digital era. The model also provides conceptual contributions to the theories of multimodal literacy and critical digital literacy.