Multimedia Web Content Generation Using Large Language Models with Chain-of-Thought Reasoning Strategy

Abstract

Recent advances in Large Language Models (LLMs) and Chain-of-Thought (CoT) reasoning have improved the quality of AI-generated multimedia content and code. However, the reliability and stability of generated outputs still depend heavily on prompt design, model capability, and application context. This study examines a human–AI collaborative framework for multimedia web content generation and programming education. By integrating CoT-guided prompting with full-stack web development processes, the framework supports frontend interface construction, backend logic implementation, and database connectivity within a unified workflow. Two prompting approaches are compared: conventional zero-shot generation and CoT reasoning. The generated outputs are evaluated using quantitative indicators, including memory usage, execution time, and test coverage, together with user-based assessments of visual aesthetics and creative expression. The results show that CoT prompting improves the logical consistency and structural completeness of AI-generated code when compared with conventional zero-shot generation. Comparative benchmarking also indicates that ChatGPT and Gemini produce more stable results than Grok in tasks involving complex algorithmic reasoning. From an educational perspective, the framework further demonstrates the potential of generative AI to support learners without strong programming backgrounds through structured prompting, intermediate feedback, and iterative refinement. These findings suggest that CoT-guided human–AI collaboration can provide practical support for multimedia web development and programming education, while also revealing the current limitations of generative AI in resource-constrained and logic-intensive tasks.

Keywords: Generative AI, Chain-of-Thought reasoning, multimedia web development, programming education, human–AI collaboration

DOI: 10.54941/ahfe1007265