The rapid mainstream acceptance of generative artificial intelligence (GenAI) creates an opportunity to strengthen student learning in teacher education. By embedding responsible GenAI use within core learning processes, such as receiving and acting on feedback for assignments and tasks, programmes can develop student teachers’ AI literacy while promoting quality learning and feedback literacy (e.g., Prompiengchai et al. 2025). In this presentation, we will present a project, funded by the Strategic Alignment of Teaching and Learning Enhancement Funding in Higher Education, aimed at developing and documenting a GenAI formative feedback agent (RebelBot) to support student teachers’ planning tasks within the professional B.Ed. Physical Education, Sports Studies, and Arts at University College Cork (UCC).

The project, currently underway, adopts a Design-Based Research-inspired (Reeves et al. 2006), multi-phase cycle of co-design, piloting, and refinement of RebelBot. Phase 1 comprised two co-creation workshops with five paid Student Partners (SPs), centring on (i) previous experience with GenAI and desired features for RebelBot and (ii) feedback structure, preferences, and guardrails.

Phase 2 comprises an iterative pilot of RebelBot, created within Microsoft Copilot Studio under UCC licensing and EU Data Boundary controls, using the OpenAI GPT-5 Reasoning model (preview), and following the overall specifications agreed in Phase 1. Four SPs completed three submission cycles of planning artefacts, each followed by a survey collecting insights on usability (User Experience Questionnaire; Schrepp et al.) and structured suggestions to inform iterative refinements. This piloting phase also includes a preliminary reliability and validity study reusing consenting students’ historic planning assignments. This seeks to evaluate (i) test–retest reliability, examining stability of agent outputs when the same artefact is processed on separate occasions under controlled prompting, and (ii) convergent validity, testing alignment between agent outputs and existing criterion-level marking by a human marker.

Phase 3 will provide cohort-wide access for all Year 2 students (n = 52) to support equitable access and will also evaluate pre–post feedback literacy (Feedback Literacy Behaviour Scale; Dawson et al.) and AI ethical learning (AI Literacy Questionnaire; Ng et al.). It will also include a reflexive focus group with lecturers (n = 3) on the perceived value and impact of the agent’s implementation.

Phase 4 will entail the creation and dissemination of materials and tutorials to support the development of similar agents across disciplines and contexts.

This presentation will summarise the design decisions, phased evaluation, and ethical and guardrail structures underpinning RebelBot, offering a transferable methodology for responsible GenAI feedback with students in accredited third-level programmes.