ASSESSING AI-SUPPORTED ACADEMIC WRITING: THRESHOLD PERFORMANCE, TRACEABILITY, AND PEDAGOGICAL MEDIATION IN A FIRST-YEAR UNIVERSITY COURSE

Background: Generative artificial intelligence (AI) has intensified debates about academic writing, authorship, and the validity of learning assessment evidence in higher education. This study examined how three pedagogical conditions of academic writing were associated with threshold performance, traceability, and pedagogical mediation in a first-year university Spanish course. Methods: A mixed-methods design was used in a general education course at the University of Puerto Rico, Ri­o Piedras Campus. The quantitative component analyzed 35 student writing performances distributed across three conditions: AI-integrated in situ writing, non-AI in situ writing, and non-AI take-home writing. Student performance was examined through an institutional analytic rubric aligned with Effective Communication, Information Literacy, and Critical Thinking. The qualitative component analyzed 12 AI writing logs to identify documented decisions of acceptance, rejection, reformulation, verification, and authorial control.