A Report on the Iterative Model and Efficacy Practice of English Speech Draft Writing from the Perspective of Human-AI Collaboration

This practice report takes the writing practice of an undergraduate student participating in an English speech contest as a case study. It examines the functional positioning of AI tools in speech draft writing and the dominant role of human writers in controlling aspects such as article style and linguistic expression. Based on a clearly defined theme and target audience, this practice employed DeepSeek and Grammarly for initial draft generation and language proofreading, followed by multiple rounds of manual review and revision to produce the final draft. The results indicate that in three key areas—structural cohesion, argumentative depth, and linguistic expression—human-led review and revision can significantly compensate for issues in AI-generated text, such as logical leaps, thin arguments, and repetitive rhetoric. Furthermore, the student demonstrated crucial value judgment and decision-making abilities in prompt design, text discrimination, and integration. This suggests that AI is better suited as a "collaborator" rather than a "replacement." Teaching and training should focus on a closed-loop process of "task definition—high-quality prompting—critical screening—integration and refinement" to cultivate students' questioning, decision-making, and leadership abilities, alongside aesthetic judgment.