Evaluating generative artificial intelligence’s role in enhancing feedback for digital multimodal compositions: a comparative study

While research has examined the use of generative artificial intelligence (GAI) in assessing monomodal writing, its potential for evaluating digital multimodal compositions remains largely unexplored. This study compares feedback generated by GAI (GPT-4o) with feedback provided by teachers on multimodal academic posters created by 88 Chinese university students learning English as a foreign language (EFL). Employing a mixed-methods design, we conducted a quantitative analysis of feedback quality across five dimensions: criteria-based assessment, clarity of directions, accuracy, prioritization of essential features, and supportive tone. Additionally, we performed a thematic analysis of teacher and student perceptions. The results indicate that GAI excelled in aligning with predefined criteria and demonstrated comparable effectiveness in maintaining a supportive tone. However, teachers outperformed GAI in providing clear directions for improvement, delivering accurate feedback, and prioritizing essential features that are both attainable and pedagogically appropriate. Teachers rated GAI feedback most favorably regarding base units and generic structure but expressed concerns about its evaluation of purpose and content. Students found teacher feedback to be more helpful and preferred it over GAI feedback, citing its specificity, relevance, and actionable nature. These findings support a complementary, role-based model of feedback in which GAI assists with consistent, rubric-aligned evaluation, while teachers focus on higher-order, context-sensitive guidance. This study contributes to the growing discourse on human-AI collaboration in multimodal writing assessment and offers practical strategies for integrating GAI into EFL writing instruction.