Against the backdrop of AI reshaping core educational mechanisms, AI has advanced personalized instruction while revealing inherent limitations in situational awareness and value trade-offs. Existing research has focused heavily on AI’s technical performance and application outcomes, yet few studies have systematically examined the roles of explainable AI (XAI) mechanisms and learners’ active questioning competence in authentic educational settings. This study addresses three core questions: how AI impacts educational effectiveness under the human-machine collaboration framework; the moderating roles of situational complexity and questioning ability; and how XAI strategies adapt to heterogeneous task contexts. Using a 3 (XAI: low/medium/high) × 2 (situational complexity: low/high) × 2 (questioning ability: high/low) between-subjects factorial design, this study collected data from 500 participants via offline experimental questionnaires, with XAI as the independent variable and educational effectiveness as the dependent variable. Results show high-level XAI significantly improved educational outcomes (F(2, 494) = 5.23, p = .006, η² = .021). XAI gradually offset complexity-induced cognitive load: complexity negatively impacted outcomes under low XAI, but this effect disappeared under high XAI. Individuals with stronger questioning competence also derived greater benefits from XAI. This study concludes that effective AI deployment in education depends on synergistic alignment of technology design, learner competence, and task context. AI functions best as a supportive tool, with efficacy tied to the coupling of explanation transparency and active questioning ability. This work provides empirical evidence for educational XAI and references for humanmachine collaborative learning design.