Retroalimentación con IA generativa en la evaluación inclusiva: Marco de mediación docente basado en el diseño universal para el aprendizaje

The incorporation of generative artificial intelligence (AI) into educational feedback offers opportunities to expand personalisation and accessibility, but it also raises risks of bias, opacity and the erosion of teacher mediation. The objective of this theoretical-conceptual article was to propose a framework to guide the use of generative AI in feedback from the standpoint of inclusive assessment and Universal Design for Learning (UDL). As method, a selective narrative review of the literature on inclusive assessment, formative feedback and AI in education was conducted across three strands, which were then integrated into a framework of teacher decisions and transversal criteria, contrasted with related models and complemented by a focused search of recent studies on trust and evaluative judgement. As a result, the framework links four teacher decisions (what is assessed, how evidence is gathered, how feedback is given, and what opportunities for improvement are offered) with three transversal quality criteria: accessibility, agency and dialogue, and equity with data protection. It is complemented by a four-level scale of teacher mediation that distinguishes lower- and higher-risk uses, a checklist for reviewing AI-supported feedback before it reaches students, three illustrative examples, and five testable propositions to guide empirical validation. The implications of the framework for teaching practice and institutional policy are discussed, along with its tensions (efficiency versus dialogicity, personalisation versus privacy, linguistic diversity, and the access gap) and its methodological limitations. The article concludes that inclusion does not depend on the tool but on the pedagogical and ethical decisions that govern its use, and that teacher responsibility for interpreting evidence remains irreplaceable.