Abstract
<title>Abstract</title> <p>This study introduces the Consistency Quotient (CQ) as a novel instrument to measure the internal coherence of AI-generated educational content. Through comparative content analysis of 356 samples from 89 teachers (38 novice, 51 experienced) across four differentiated models—Simplified, Enriched, Integrated Multi-Level, and Mixed-Ability—we examined how professional experience shapes coherence across conceptual, linguistic, and structural dimensions. Experienced teachers achieved significantly higher overall CQ scores (M = 73.8 vs. 67.4, d = 0.62), with the largest advantages in conceptual (d = 0.79) and structural coherence (d = 0.77). Linguistic coherence showed a non-significant difference (d = 0.38), suggesting that structured prompt engineering effectively supports linguistic adaptation regardless of experience. Critically, negative case analysis revealed that 18.4% of novice teachers outperformed the experienced mean, while 11.8% of experienced teachers underperformed the novice mean, and overlap analysis indicated a 38.2% distribution overlap—challenging deterministic assumptions about professional experience. Pairwise comparisons identified the Simplified-Enriched pair as the most coherent and the Simplified-Mixed pair as the least coherent, highlighting the distinct challenges of vertical versus horizontal differentiation. These findings underscore that AI literacy and enacted pedagogical content knowledge are pivotal determinants of coherence, extending AI-TPACK frameworks by focusing on artifactual quality. The CQ offers a robust, practically applicable instrument for assessing teacher-AI collaboration and informing targeted professional development.</p>