The Future of Education

The growing diffusion of generative artificial intelligence (AI) in school contexts has intensified pedagogical reflection on formative assessment and feedback processes, particularly within assessment for learning frameworks [1,2,3]. Recent studies highlight the potential of AI to enhance the timeliness, consistency and sustainability of feedback practices in educational settings [4,5], yet empirical evidence in compulsory education remains limited, especially in contexts characterised by teacher supervision and criterion-referenced assessment. This study investigates the effectiveness, clarity and pedagogical sustainability of AI-generated feedback compared with teacher-provided feedback in primary and lower secondary education. The research was conducted during the 2024/2025 school year in two comprehensive schools in Northern Italy and involved 100 students across seven classes and five teachers from different subject areas. Adopting a mixed-method empirical–comparative design, the study integrated quantitative analyses of learning outcomes with survey data on students’ perceptions. Written curricular tasks were assessed using shared analytic rubrics, and feedback was generated either by teachers or by a generative AI system (GPT-4o), guided by carefully designed prompts aligned with the same assessment criteria [6]. After receiving feedback, students revised their work, enabling pre–post comparisons of performance. Results show statistically significant improvements in both groups, with higher average learning gains associated with AI-generated feedback in five out of seven classes. High levels of agreement between human and automated assessments were observed (ICC > .92), indicating strong consistency and reliability. In line with research on formative feedback and feedback literacy [7], students reported high clarity and usefulness of feedback regardless of its source and demonstrated strong engagement in revision processes. Overall, the findings suggest that AI-generated feedback, when embedded within a pedagogically intentional and teacher-supervised framework, can effectively support assessment for learning by enhancing timeliness, coherence and inclusivity, while complementing rather than replacing teachers’ professional judgment.