Designing AI-generated multi-format reading assessment across Bloom’s cognitive levels: effects on L2 reading comprehension, cognitive engagement, and motivation

Artificial intelligence (AI) has advanced automated question generation, yet implementations remain single-format and English-dominant. This study examined an AI-generated multi-format reading assessment platform for Chinese-speaking learners of Indonesian as a second language (L2), focusing on reading comprehension, motivation, and cognitive engagement. The platform generated eight formats across Bloom’s revised taxonomy levels (LOTS, MOTS, HOTS) and integrated a trilingual interface, retrieval-augmented chatbot, multiplayer assessment, learning analytics, instructor dashboard, and accessibility tools: text-to-speech, display settings, and visual filters. A quasi-experimental mixed-methods design involved 300 learners (n = 150 experimental; n = 150 control) at three universities over 12 weeks. Experimental learners used the integrated platform; controls read identical texts with multiple-choice-only quizzes. ANCOVA showed higher post-test comprehension, F(1, 297) = 26.14, p < .001, partial η 2 = .08; corrected contrasts showed larger effects at LOTS, MOTS, and HOTS (d = 0.35, 0.54, 0.89). Engagement increased, F(3.72, 554.28) = 18.67, p < .001, partial η 2 = .11, and motivation was higher, t(298) = 5.18, p < .001, d = 0.60. Interviews with 24 learners showed format diversity, taxonomy-based scaffolding, contextual chatbot support, gamification, and accessibility shaped usefulness and engagement. Findings support AI-generated multi-format assessment and Bloom-based item generation for less commonly taught L2s.