Large Language Models (LLMs) such as ChatGPT are increasingly used in education, but their general-purpose design limits support for conceptual learning. We present a study of a custom GPT, scaffolded with Bloom’s Taxonomy and framed as an educator, to examine how structured dialogue shapes learning. Using deceptive patterns as a test domain, 38 students compared the custom GPT with ChatGPT-4 and an expert-authored website. The custom GPT led to higher satisfaction, trust, and perceived learning than ChatGPT-4, though the website remained most trusted overall. Students valued structured guidance for deeper reflection but also found it demanding, and still relied on expert sources for credibility.
‹ All reading
From ChatGPT to Custom GPTs: Scaffolding Conceptual Learning in Adult Education
A study of a custom GPT, scaffolded with Bloom’s Taxonomy and framed as an educator, to examine how structured dialogue shapes learning, and proposes design directions that combine scaffolding with source cues, adapt to learner needs, and complement teaching practices.