Student Engagement and Generative AI-Mediated Written Feedback in Primary and EarlyChildhood Preservice Teacher Education

The EFAI project examines the impact of generative AImediated written corrective feedback (WCF) compared to human feedback on the development of academic writing competence and the engagement of preservice early-childhood and primary teachers. WCF is a key pedagogical resource, yet its effectiveness remains debated due to variations in student characteristics, task features and feedback types. The emergence of generative AI introduces new possibilities for personalized, rapid and iterative feedback, while also raising concerns regarding accuracy, biases, transparency and data privacy. EFAI aims to address these challenges by analysing how AI-generated feedback differs from human feedback, how students interact with each type, and how these interactions influence their learning processes and emotional responses. The project adopts a longitudinal quasi-experimental design with two groups (human vs. AI feedback) to compare the clarity, specificity and explanatory depth of feedback; the improvement of spelling, punctuation, anaphoric demonstrative deictic devices (esto/eso/aquello) and discourse coherence; and the behavioural, cognitive and affective dimensions of student engagement. Data sources include initial and revised texts, digital traces, IAGstudent interactions, mini-surveys, reflective chats, interviews and focus groups. A custom-built AI tool will generate direct, indirect and metalinguistic feedback using randomized allocation. The methodological approach integrates quantitative analyses (prepost comparisons, improvement indices, SEM, multilevel modelling) with qualitative analyses (thematic coding, blind text annotation and process-oriented examination of revisions). EFAI will produce original insights into how feedback quality shapes revision strategies, how students perceive and trust AI-generated comments and how emotional factors influence the uptake of feedback. It will also explore shifts in university teachers roles when AI reduces repetitive correction tasks. Expected outcomes include theoretical advancements in understanding engagement with WCF, evidence-based guidelines for the pedagogical and ethical use of generative AI, and the creation of open-access materials and validated instruments. The project contributes to broader social goals by strengthening preservice teachers writing skills, reducing educational inequalities through accessible resources and fostering critical AI literacy aligned with UNESCO recommendations and SDG priorities. EFAI will strengthen an international research network and culminate in high-impact publications, a specialized conference and an open research dataset. Impacto científico técnico o internacional esperable: The EFAI project will generate significant scientific and technical impact by addressing an emerging and still underexplored area: the use of generative artificial intelligence (Gen AI) to provide written corrective feedback (WCF) in initial teacher education. Although WCF is a central component in the development of writing competence, its effectiveness remains a matter of debate, particularly regarding the types of feedback, the incorporation of corrections, and learners individual differences. The rise of Gen AI introduces new possibilitiespersonalization, immediacy, automation of repetitive tasksbut also methodological and ethical challenges related to feedback quality, model reliability, algorithmic bias, and data privacy. In this context, EFAI will provide rigorous empirical evidence that helps delineate the actual role of Gen AI in teaching and learning processes of academic writing. The combination of a longitudinal quasi-experimental design, advanced quantitative analyses (SEM, multilevel models), and in-depth qualitative approaches (thematic analysis, blind coding, revision-process review) will generate original knowledge on three key dimensions of writing development: the nature and quality of feedback generated by Gen AI and by humans; the effects of both on improving discourse coherence; and the relationship between these types of feedback and the behavioral, cognitive, and affective dimensions of student engagement. The use of an ad hoc Gen AI toolcapable of randomly controlling the feedback modality (direct, indirect, and metalinguistic) and recording digital interaction tracesconstitutes a novel methodological contribution to research in academic writing and educational technology. EFAI will advance theoretical frameworks on written feedback and student engagement introducing models that integrate cognitive attention, academic emotions, revision operations, and learners decision-making in hybrid humanAI environments. Likewise, the project will identify usage conditions, potential benefits, and limitations of Gen AI in authentic higher education contexts, offering quality criteria applicable to other disciplines and training settings. At a technical level, it will provide replicable protocols for the design, evaluation, and auditing of Gen AI tools for teaching, promoting standards of transparency, traceability, and experimental control of automated feedback. The impact will be reflected in tangible contributions to the scientific community: publications in high-impact journals, an open repository of data and resources, the creation of a large anonymized corpus of texts with human and Gen AI feedback, and the consolidation of an international research network specializing in WCF and educational AI. Taken together, EFAI will substantially advance knowledge about how future teachers learn to write, how they interact with different sources of feedback, and how Gen AI can be integrated ethically, effectively, and evidence-based into higher education.