How YouTube Frames ChatGPT Use in Education: An Epistemic Network Analysis with Supporting Multimodal Metadata

Synthesis: Uses epistemic network analysis of Multimodal AI YouTube metadata (transcripts, titles, thumbnails, comments) to show how different creator groups frame ChatGPT use in education, revealing divergent narratives around learning support versus academic-integrity risk. The work connects to broader debates about how Generative AI systems reshape Student Experience and the conditions under which AI support scaffolds rather than undermines learning. It has direct implications for The Pedagogy of AI Mistakes: Fostering Higher-Order Thinking and the risk of Over-Reliance when assistants absorb too much of the cognitive load. Findings also bear on AI Literacy and Self-Regulated Learning, and on how institutions should govern Student Experience and Academic Integrity. Practitioners in Higher Education and teachers can use the evidence to calibrate when to deploy Large Language Models (LLMs)-based help and how to pair it with Feedback that preserves learning gains.

What this means for practice

• Instructors. Separate the three framings before you assign AI use: conceptual scaffold, retrieval practice, and productivity substitute are not interchangeable, and the analysis found three structurally distinct discourse groups across 52 videos and 557 coded transcript chunks.
• Instructors. Teach with the retrieval-practice framing, which aligned with active cognitive engagement, rather than the productivity framing, which positioned ChatGPT as a substitute for effort.
• Learners. Read "use ChatGPT to finish faster" videos as advice about output, not about learning: productivity-framed content framed the tool as replacing effort yet reached engagement comparable to skill-oriented content.
• Learners. Seek out learning-aligned videos deliberately rather than trusting popularity as a quality signal, since the most learning-aligned content had the lowest visibility while skill- and productivity-oriented content reached substantially more learners.

Limitations

• The study coded creator transcripts, titles, thumbnails, and viewer comments rather than measuring learning outcomes, so it cannot show what viewers actually learned.
• Comments were unevenly distributed across groups, with most highly liked productivity-group comments coming from a single video, which limits group comparisons.
• The 52 videos were collected under one search strategy and a fixed time period, so the findings may not generalize to all LLM-related educational content.
• Group classification and the epistemic network analysis used the same nine codes, so the ENA describes co-occurrence patterns within predefined groups rather than independently confirming the group separation, and the coding scheme reflects the researchers' theoretical perspective.

Citation

L. Xiao, G. Chen, Y. Zhang et al. (2026). How YouTube Frames ChatGPT Use in Education: An Epistemic Network Analysis with Supporting Multimodal Metadata.