Surrey researchers argue for care and human relationships in AI-supported feedback

Updated Sep 07, 2026

A University of Surrey release published on 31 March 2026 describes an argument for preserving care, trust and relationships as generative AI becomes part of educational feedback. It highlights both possible complementary uses and risks of reducing feedback to automated comments. Surrey release.

The contribution and its limits

The paper is The care-full craft of feedback in an age of generative AI, by Naomi E. Winstone and colleagues, published online on 18 March 2026 in Assessment & Evaluation in Higher Education. Its DOI is 10.1080/02602938.2026.2643333.

The accessible publisher text presents a conceptual discussion drawing on existing scholarship and a feedback manifesto. It is not a newly reported experiment comparing student learning with human and AI feedback. Its framing emphasises meaning-making, educational relationships, trust and the professional work involved in feedback. Publisher article.

Surrey summarises ten manifesto principles, including designing feedback with learners and educators and prioritising learning over technological efficiency. Statements about student trust in human feedback refer to the literature discussed, rather than a new survey conducted for this paper. This review has checked the release and accessible publisher passages, not independently reanalysed those underlying studies.

Questions for an institutional pilot

Our practical suggestion is to define the educational purpose before choosing an AI workflow. What should students understand or do after receiving feedback? Who can explain or challenge it? How will the institution check whether that process actually occurred?

Ask students about usefulness, clarity and trust separately. An answer that feels reassuring can still be inaccurate; a challenging response can still support learning. Combine their accounts with appropriate checks of feedback accuracy and use, rather than treating satisfaction as the learning outcome.

Agree where professional judgement remains necessary and how students can seek a human discussion. Consider access, confidentiality and different support needs when designing any evaluation. The authors’ argument can inform these choices, but it does not validate a particular AI product, guarantee a learning benefit or create a regulatory requirement.

The student comment analysis governance checklist offers questions for planning how pilot feedback will be collected, accessed and interpreted.

Review note, 7 September 2026: verified the paper’s identity and online publication date, identified its conceptual contribution, and distinguished authors’ arguments and cited literature from new empirical findings. The summary is bounded to the accessible release and publisher text.