國小中年級自然科學習逐漸由生活經驗的觀察，轉向較具學科性的概念理解。學生在學習過程中，不只需要讀懂文本，也需要整理資訊、釐清概念，並以自己的話表達想法。然而，在教學現場中，部分學生閱讀多篇自然科文本時，容易出現能理解部分內容卻不易整理、知道一些內容卻無法完整表達，或不知道如何提問與作答的情形。基於此，本研究以國小四年級自然科 AI 輔助深讀活動為情境，探討學生在多文本閱讀與生成式 AI 輔助歷程中的學習經驗。 本研究採質性描述研究取向，以國小四年級三個班級共93名學生為研究對象，於112學年度第二學期實施自然科多文本深讀活動，並導入 ChatGPT 作為學習輔助工具。學生透過明日星球平台進行自然科多文本閱讀、個人讀寫與小組討論，並在教師引導下使用 ChatGPT 進行概念釐清、資訊查詢、重點整理與提問延伸。研究以76份有效開放式問卷為主要分析資料，並參照明日星球平台作答紀錄與189筆可辨識的 ChatGPT 互動事件。資料分析經由資料整理、反覆閱讀、初步編碼、類別整併與主題歸納，整理學生的 AI 使用方式、學習感受、學習困難與支持需求。 研究發現，學生多在閱讀或作答時遇到不理解的內容、不知道如何回答，或需要即時協助時使用 ChatGPT。從學生自述來看，其使用方式主要包括詢問自然科概念、查詢答案或相關資訊、整理重點與參考作答；實際互動事件則以概念釐清與答案查詢為主，兩者合計占75.7%，延伸追問、應用推論與資訊整理所占比例相對較低。多數學生自述 ChatGPT 能提供概念說明與作答參考，使其在理解或作答遇到困難時較容易獲得協助；部分學生也因 AI 回應中的補充說明或延伸內容，而對自然科主題產生好奇。不過，學生在自然科深讀歷程中仍面臨閱讀理解、資訊整理、概念掌握、問題形成與作答表達等困難，也不一定能判斷 AI 回應是否適切或正確。 整體而言，生成式 AI 可作為自然科深讀歷程中的即時輔助資源，但其價值不在於取代閱讀、思考或教師引導，而在於能否被納入適切的學習歷程。自然科 AI 輔助深讀的關鍵，在於多文本閱讀、教師引導、同儕互動、閱讀支架與 AI 輔助之間的整合。教師若能提供提問示範、文本查證與資訊整理支架，較能協助學生檢視並整理 AI 回應，進而轉化為自己的理解與表達。;As elementary school students progress into the middle grades， science learning gradually shifts from observations based on everyday experiences toward the understanding of more subject-specific concepts. Students need not only to comprehend science texts but also to organize information， clarify concepts， and express ideas in their own words. In classroom practice， however， some students may understand only part of the content but have difficulty organizing information across multiple science texts， expressing what they know clearly， or formulating questions and written responses. Against this background， this study explored fourth-grade students’ learning experiences in AI-assisted deep reading activities in elementary science， focusing on multiple-text reading and the use of generative AI as a learning support tool. This study adopted a qualitative descriptive approach and involved 93 fourth-grade students from three classes in an elementary school. The activities were implemented during the second semester of the 2023–2024 academic year in Taiwan. Students used the Planet of Tomorrow platform to read science texts， complete individual reading and writing tasks， and participate in group discussions. Under teacher guidance， they also used ChatGPT to clarify concepts， search for information， organize key points， and extend their questions. The primary data consisted of 76 valid responses to an open-ended questionnaire， supplemented by platform response records and 189 identifiable ChatGPT interaction events. The data were analyzed through data organization， repeated reading， initial coding， category consolidation， and thematic synthesis. The findings showed that students commonly used ChatGPT when they encountered content they did not understand， were unsure how to answer a question， or needed immediate assistance. According to their self-reports， students used ChatGPT to ask about science concepts， search for answers or related information， organize key points， and obtain guidance when constructing responses. The interaction records showed that concept clarification and answer seeking accounted for 75.7% of the identifiable interactions， whereas extended questioning， application and inference， and information organization occurred less frequently. Most students reported that ChatGPT provided conceptual explanations and guidance for constructing responses， particularly when they experienced difficulty understanding texts or answering questions. Some students also expressed greater curiosity about science topics after receiving additional explanations or related information from AI. Nevertheless， students continued to experience difficulties in reading comprehension， information organization， concept understanding， question formulation， and written expression. They did not always know how to formulate clear questions or determine whether AI-generated responses were appropriate or accurate. Overall， generative AI may serve as a source of immediate support during deep reading in elementary science. Its value， however， does not lie in replacing reading， thinking， or teacher guidance， but in how appropriately it is integrated into the learning process. The effective use of generative AI in deep reading depends on the integration of multiple-text reading， teacher guidance， peer interaction， reading scaffolds， and AI support. When teachers model question formulation and provide scaffolds for text-based verification and information organization， students may be better able to evaluate， organize， and integrate AI-generated information into their own understanding and written responses.
