Michael Henderson, Margaret Bearman, Jennifer Chung, Tim Fawns, Simon Buckingham Shum, Kelly E. Matthews & Jimena de Mello Heredia (2026) — Assessment & Evaluation in Higher Education 51(5), 863–878. doi:10.1080/02602938.2025.2502582.
📄 Full text (Taylor & Francis, OA)
Summary
The largest study in the AEHE 51(5) special issue: a cross-sectional survey across four Australian universities (≈192,000 invited; 10,132 volunteered; this paper analyses 6,960 students who answered the feedback items). It combines quantitative comparison of perceived helpfulness/trustworthiness of GenAI vs teacher feedback with thematic analysis of 8,642 open-ended responses (11,903 coded instances, 48 codes). Core conclusion: GenAI and teacher feedback serve different needs — complementary but not interchangeable.^[raw/papers/tandf-2026-genai-teacher-feedback-comparison.md]
Quantitative findings
- Usage split: 49.7% (n = 3,461) of students had sought feedback from GenAI; 50.3% had not.
- Helpfulness: 83.9% rated AI feedback somewhat/very helpful vs 82.2% for teacher feedback — a small but significant teacher advantage (mean diff 0.14, t(3327) = −7.17, p < .001, d = −0.13).
- Trustworthiness: the striking gap — 90.5% rated teacher feedback somewhat/very trustworthy vs 60.1% for AI (mean diff 0.93, t(3327) = −49.30, p < .001, d = −0.89, a large effect). 58.3% rated teacher feedback very trustworthy vs 8.8% for AI.^[raw/papers/tandf-2026-genai-teacher-feedback-comparison.md]
Qualitative findings — why they differ
When comparing GenAI to teacher feedback, students said GenAI was more: accessible/easy (99.3% of access codes), fast, voluminous, understandable, objective (less biased), and positive in tone — and less risky (99.3% of relational-risk codes: less vulnerable, no loss of social status). It aided sense-making and could be used before submission.
Teacher feedback was more: relevant, contextualised (95.2% of contextualisation codes), specific, in-depth, personal/relational, and expert — but also more frequently negative in tone (e.g. dismissive, insulting) and more likely to produce negative feelings (85.3% of negativity codes).^[raw/papers/tandf-2026-genai-teacher-feedback-comparison.md]
The two most frequent comparison themes were the nature of the feedback information (54.8% of coded instances; quality, reliability, relevance, contextualisation) and feedback processes (22.7%; access, timing, effort, sense-making).
Why students did NOT use GenAI for feedback (n = 3,405 comments)
- 28.1% were unaware it was possible or did not know how (a support/equity gap, not a preference).
- 28.7% cited trustworthiness/reliability concerns.
- 22.5% cited values: preference for human connection (4.3%) or no perceived need (6.3%).
- Smaller shares: academic integrity (9.1%), privacy (2.7%), wanting to preserve effortful learning (18 of 24 effort codes).^[raw/papers/tandf-2026-genai-teacher-feedback-comparison.md]
Implications
- Students already self-initiate GenAI feedback at scale, so institutions should actively support how students engage with it (feedback literacy, evaluative judgement — cf. ai-literacy).
- The trust gap (90.5% vs 60.1%) is not simply an accuracy verdict; it partly reflects source-credibility heuristics (Lipnevich & Smith 2008; Nazaretsky et al. 2024) and may be context-dependent (language-form advice vs deeply contextual course guidance).
- The "less risky" property makes GenAI feedback valuable for feedback seeking, especially for anxious students — complementing chatgpt-feedback-engagement-genai's "calm, stress-free" affective finding — while teacher feedback retains contextual expertise and relational recognition.^[raw/papers/tandf-2026-genai-teacher-feedback-comparison.md]
- Directly challenges replacement narratives: GenAI is an additional source, not a substitute for the teacher-learner relationship (cf. ai-generated-feedback-higher-ed, which found equivalent outcomes under strong assessment architecture — outcomes and perceptions can diverge).
Related Pages
- feedback-futures-genai — Editorial synthesis; this study anchors the usefulness–trust–uptake tension
- chatgpt-feedback-engagement-genai — Deep engagement study: trust split, affective calm, uptake rates
- learner-centered-feedback-ai — Teacher-side interaction with AI feedback tools
- ai-generated-feedback-higher-ed — Randomised equivalence study: AI vs teacher feedback outcomes
- ai-feedback-quality — Quality and validity of AI-generated feedback
- feedback-loop — Feedback as process, not information
- student-experience — Student perceptions and affective responses
- ai-literacy — The evaluative-judgement capacities students need
- higher-ed — Deployment context (four Australian universities)
- repeated-ai-writing-feedback-semester — Semester-long student evaluation of repeated GenAI writing feedback
Citation
APA: Henderson, M., Bearman, M., Chung, J., Fawns, T., Buckingham Shum, S., Matthews, K. E., & de Mello Heredia, J. (2026). Comparing Generative AI and teacher feedback: Student perceptions of usefulness and trustworthiness. Assessment & Evaluation in Higher Education, 51(5), 863–878. https://doi.org/10.1080/02602938.2025.2502582