近年生成式AI逐漸進入高等教育研究與學習情境，學術圖書館也開始推動AI素養。本研究以政大圖書館三類AI推廣活動為場域，採便利取樣取得263份有效問卷，以14題知識檢測、3題能力自評及3題開放作答描繪參與者的AI素養輪廓。結果顯示，知識檢測整體答對率為79.5%，以AI倫理與應用AI較佳，理解AI相對較弱；能力自評平均為3.87分。量化與開放作答共同顯示，參與者已將AI廣泛用於資料整理、摘要、資訊查找、翻譯及研究寫作，並具有查證、隱私與責任等風險意識，且不同身分與學生學科群組呈現AI知識表現與能力自評未必一致的現象。三類首次填答來源亦觸及不同受眾展現不同AI素養輪廓，圖書館可透過多元活動形式互補推廣AI素養。

Generative artificial intelligence (AI) is increasingly integrated into research and learning in higher education, creating new needs for academic libraries to support AI literacy. This study examines the AI literacy profiles of participants in three types of AI outreach activities organized by National Chengchi University Library: a prize quiz, database workshops, and an AI-themed lecture series. Using convenience sampling, 263 valid questionnaires were collected. The questionnaire included 14 knowledge questions selected from the Generative AI Literacy Assessment Test (GLAT), three self-evaluation items adapted from AILIS 1.0, and three open-ended questions on AI use, limitations, and possible future applications. The overall correct response rate for the knowledge questions was 79.5%. Participants performed better in AI ethics (81.7%) and applying AI (81.4%), while understanding AI showed the lowest correct response rate (76.0%). The mean score for the three self-evaluation items was 3.87 out of 5. Quantitative and open-ended results together showed that participants had already incorporated AI into information organization, summarization, information and literature searching, translation, and academic or other writing. They also demonstrated strong awareness of verification, hallucination, privacy, originality, and responsibility, while discussion of underlying technical mechanisms was less common. Importantly, extensive AI use did not imply unconditional acceptance: the same task could be considered appropriate for AI assistance but inappropriate for full delegation depending on accuracy, verifiability, professional judgment, and responsibility. Knowledge performance and self-evaluated ability did not always follow the same pattern across status and disciplinary groups. The three initial response sources also reached different participant groups, suggesting complementary roles for different outreach formats. Because of convenience sampling, unequal group sizes, and the absence of pre- and post-tests, the findings should be interpreted descriptively rather than as population differences or activity effects. The results suggest that academic libraries should combine practical AI use with technical understanding, source evaluation, disciplinary context, risk judgment, and human responsibility.

DOI: 10.6575/JILA.202606_(108).0002