本研究以行動研究（Action Research）為方法論，以教師即研究者（teacher-as-researcher）的身分，在台灣北部某公立高中高三英文寫作課堂中，系統性地實施、觀察、反思並調整三個連續的教學循環，旨在增進研究者本身在知識翻新（Knowledge Building， KB）理念與 AI 工具（ChatGPT）融合教學方面的專業知能，特別是如何引導 EFL 學生在有工具支援的學習情境中真正發展英文寫作能力，而非依賴工具代替思考。 研究對象為研究者任教班級之 30 位高三學生，研究歷程橫跨一學期，涵蓋四次全校模擬考與大學學科能力測驗（學測）。研究者在每一教學循環結束後整理分析問卷、課堂觀察、學生作品、AI 互動紀錄（Action Log）、半結構式訪談及量化成績等六類資料，作為下一循環教學設計與實務改進的依據，並在三個循環結束後，從教師在鷹架設計、AI 使用引導與學生思考歷程可見化三個面向的改變歷程，提出具體的教學反思與建議。 三個循環各有核心轉變：第一循環著重於建立「想法生成—社群互動—AI 協作—寫作實踐—修訂改進」的基本學習框架，啟動學生參與；第二循環在觀察到學生修訂多停留於語言層面後，透過「觀點—理由—理據」結構化框架與 Action Log 修訂決策紀錄，嘗試讓學生的思考歷程變得可見；第三循環則因意識到鷹架若長期存在反而可能阻礙能力內化，而系統性地撤除 AI 工具用量與結構性提示，引導學生進行跨循環的後設認知整合，以檢驗所學是否真正轉化為個人能力。 本行動研究的結論，是教學者在三個循環中得到的實務體會：AI 工具在課堂中發揮最大價值的方式，不是替學生完成文章，而是作為「思考觸發工具」，在適當的引導設計下促使學生比較、判斷、修正自己的想法。藉由這次行動研究，研究者除了發現「鷹架何時撤除」遠比「鷹架如何建構」更難、也更關鍵，也建議教師在規劃 AI 融入寫作教學時，應將「建構—強化—撤除」視為完整的動態設計邏輯，而非只停在讓學生把作文寫好。;This study adopts action research as its methodology， with the researcher assuming the role of teacher-as-researcher in a senior English writing course at a public high school in northern Taiwan. Over three consecutive teaching cycles， the researcher systematically implemented， observed， reflected upon， and refined instructional practices with the aim of enhancing professional competence in integrating Knowledge Building (KB) principles with AI tools (ChatGPT). A central concern was how to guide EFL students toward genuinely developing English writing ability within a tool-supported learning environment， rather than relying on tools as a substitute for thinking. The participants were 30 twelfth-grade students from the researcher′s own class. The study spanned one full semester and encompassed four school-wide mock examinations as well as the General Scholastic Ability Test (GSAT). At the conclusion of each teaching cycle， the researcher collected and analyzed six categories of data — questionnaires， classroom observations， student writing samples， AI interaction records (Action Logs)， semi-structured interviews， and quantitative test scores — to inform the instructional design and practical adjustments of the subsequent cycle. Upon completing all three cycles， the researcher synthesized findings across three dimensions of change: scaffold design， guidance on AI use， and the visualization of students′ thinking processes， ultimately offering concrete pedagogical reflections and recommendations. Each cycle was defined by a core transformation. The first cycle focused on establishing a foundational learning framework — idea generation， community interaction， AI collaboration， writing practice， and revision — in order to initiate student engagement. The second cycle， prompted by the observation that students′ revisions tended to remain at the surface level of language， introduced a structured "Claim–Reason–Evidence" framework alongside Action Log revision-decision records in an attempt to make students′ thinking processes visible. The third cycle， driven by the recognition that prolonged scaffolding may impede the internalization of competence， systematically reduced both AI tool usage and structural prompts， guiding students through cross-cycle metacognitive integration to assess whether learning had genuinely been transferred into personal capability. The study concludes with the practical insights the researcher gained across the three cycles: AI tools realize their greatest value in the classroom not by completing essays on students′ behalf， but by serving as thinking-triggering tools — prompting students， through well-designed guidance， to compare， evaluate， and refine their own ideas. Through this action research process， the researcher also discovered that knowing when to remove scaffolding is far more challenging — and far more consequential — than knowing how to construct it. Teachers planning to integrate AI into writing instruction are accordingly encouraged to treat "construction–reinforcement–removal" as a complete and dynamic design logic， rather than stopping at simply helping students produce better essays.
