The DOI matches its Crossref record; 48 of the 48 references it cites were found in a registry or an index; 4 of the 15 sources its claims cite were checked for support.
Comparing Student Preferences for AI-Generated and Peer-Generated Feedback in AI-driven Formative Peer Assessment
Open accessAbstract
Formative assessment can enhance student learning and improve teaching practices by identifying areas for growth and providing feedback. However, practical obstacles remain, such as time constraints and students’ passive participation and the low quality of peer feedback. Artificial intelligence (AI) has been explored for its potential to automate grading and provide timely feedback, making it a valuable tool in formative assessment. Nevertheless, there is still limited research on how AI can be used effectively in the context of formative peer assessment. In this study, we conducted an AI-driven formative peer assessment with 108 secondary school students. During the peer assessment process, students not only evaluated peers’ responses and received peer-generated feedback, but also evaluated AI-generated responses and received AI-generated feedback. This research focused on analyzing the differences in preference between AI-generated and peer-generated feedback using trace data and dispositional data. In scenarios where student participation was low or the quality of peer feedback was insufficient, students showed a higher preference for AI-generated feedback, demonstrating its potential utility. However, students with high Math Confidence and AI Interest preferred peer-generated feedback. Based on these findings, we will propose practical strategies for implementing AI-driven formative peer assessment.
Overview
CoChat Summary
TL;DR
In an AI-driven formative peer assessment with 108 South Korean secondary school sophomores, students in lower-achievement classes preferred AI-generated feedback, while a high-achievement/high-interest class preferred peer-generated feedback. Dispositional factors and trace data predicted feedback preferences and informed implementation recommendations.
Key findings
- Class A and Class C preferred AI-generated feedback over peer-generated feedback (ANOVA significant at p<0.0001 and p<0.05, respectively); Class D preferred peer-generated over AI-generated feedback (p<0.01); Class B showed no statistically significant preference for peer-generated over AI-generated feedback.
- Disposition-level comparisons found that low Math Confidence, low Math Value, low AI Relevance, and low AI Interest groups preferred AI-generated feedback over feedback from peers who answered incorrectly on their first attempt (Group 3) at p<0.1, p<0.01, p<0.1, and p<0.1, respectively; Math Anxiety and Math Interest showed no statistically significant differences.
- AI-generated feedback preference model predictors: Math Confidence −0.277 (p<0.01), Math Value −0.165 (p<0.1), PeerPreferences 0.855 (p<0.001), CorrectRatio −0.205 (p<0.05). Peer-generated feedback preference model predictors: Math Anxiety −0.133 (p<0.10), Math Value −0.176 (p<0.10), AI Relevance −0.160 (p<0.05), Math Interest 0.252 (p<0.01), Feedback Length 0.194 (p<0.01).
- Recommended practices include using AI-generated feedback in low-participation classes, math-learning difficulties, and low correct-response-rate problems; ensuring validity because high AI Interest students are sensitive to hallucination; and prioritizing human peers for students with high Math Confidence and Math Value.
Methods
- Participants were 108 secondary school sophomores in four mathematics classes (Class A N=27, Class B N=26, Class C N=29, Class D N=26) in South Korea; the course was face-to-face, with 2–3 AI-driven formative peer assessments per class using the same problems, and AI (GPT-4) participated anonymously as a peer.
- Students submitted a solution and answer, evaluated peers’ and AI-generated responses, received anonymous AI- and peer-generated feedback, and rated feedback effectiveness on a 5-point Likert scale. Trace data included FirstAttemptCorrect, FirstAttemptTime, Preference, FeedbackLength, PeerPreferences, PeerFeedbackLength, and CorrectRatio; dispositional survey data measured Math Anxiety, Math Confidence, Math Interest, Math Value, AI Relevance, and AI Interest after confirmatory factor analysis item removal (α = 0.84, 0.91, 0.75, 0.85, 0.81, 0.85, respectively).
- Analysis used ANOVA with Dunnett’s post-hoc test to compare preferences among Group 1 (AI-generated, N=182), Group 2 (peer correct first attempt, N=153), and Group 3 (peer incorrect first attempt, N=154); linear regression predicted feedback preference from trace and dispositional data after a Linear Mixed Model random effect was not statistically significant.
Limitations
- Stated: the study was conducted face-to-face with 108 students; future work should explore distance learning, higher education, and MOOCs. Dispositional data scope was limited, and future research should include self-regulated learning, goal-setting, help-seeking, and motivation. AI hallucination rate varied across items, and future work could optimize the model. AI facilitated evaluation despite low participation but did not increase participation itself.
- Evident from text: the sample is limited to secondary school sophomores in one mathematics course in South Korea, so findings may not generalize to other populations or subjects.
- Evident from text: the provided text omits numeric R² values for the prediction models, so variance explained cannot be assessed from this text.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
3 Valid · 3 Not valid · 10 Not checked · 72 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In Class C and Class D, the differences were significant at the 0.05 and 0.01 significance levels, respectively.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In Class C and Class D, the differences were significant at the 0.05 and 0.01 significance levels, respectively.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Finally, the CorrectRatio coefficient was − 0.205, revealing a significant negative association at 0.05 significance level.”
§ 4 Results > 4.3 Prediction Model of AI-Generated and Peer-Generated Feedback Preference
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“ANOVA analysis of Class C also demonstrated a statistically significant preference for AI-generated feedback (Group 1) over peer-generated feedback (Groups 2 and Group 3).”
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper states — it cites no source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In contrast, students in Class C were less likely to detect and criticize AI-generated feedback compared to Class D.”
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
44 more claims
Create a free account to see all 88 claims and to verify the 3 references still open.
- Probably not a substantial claim. 20
- It may state two separate claims as one. 26
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“(2) develop their ability to interpret, discriminate, and critically select among feedback from AI, teachers, and peers [128-130];”
From A Systematic Review of Generative AI in K–12: Mapping Goals, Activities, Roles, and Outcomes via the 3P Model (2025) · cites this paper as [129]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Some studies also integrate social-emotional and interdisciplinary tasks, where AI provides focused evaluations and follow-up recommendations that help students refine learning strategies and strengthen self-regulation [121,129,151].”
From A Systematic Review of Generative AI in K–12: Mapping Goals, Activities, Roles, and Outcomes via the 3P Model (2025) · cites this paper as [129]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Learner differences-cultural background, personal interests [129,132]-and concerns about academic integrity [85] further shape uptake.”
From A Systematic Review of Generative AI in K–12: Mapping Goals, Activities, Roles, and Outcomes via the 3P Model (2025) · cites this paper as [129]
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 1 Introduction
“Since the assessment was conducted anonymously, students could not determine who provided the feedback, whether it was from AI or a peer.”
§ 5 Discussion on Implementation Practices
“students, due to their high AI Interest, were able to easily identify AI-generated feedback among the anonymous responses”
Citations
Cites
Hattie, J., & Timperley, H. (2007) · Review of Educational Research · doi:10.3102/003465430298487 · Cited by 12,905 · cited 2×
“Also, we instructed the AI to deliver feedback at both the task and process levels, following the four levels of effective feedback proposed by Hattie [17].”
Assessment and Classroom Learning
Black, P., & Wiliam, D. (1998) · Assessment in Education: Principles, Policy & Practice · doi:10.1080/0969595980050102 · Cited by 8,152 · cited 1×
“In the current literature, formative assessment encompasses all activities conducted by teachers, learners, and peers, from which evidence is elicited, interpreted, and used as feedback to adjust their instruction and learning [6, 8].”
Expectancy-value theory of achievement motivation
Wigfield, A., & Eccles, J. S. (2000) · Contemporary Educational Psychology · doi:10.1006/ceps.1999.1015 · Cited by 7,383 · cited 1×
“Math Interest was measured with 2 items from [46].”
Formative assessment and the design of instructional systems
Sadler, D. R. (1989) · Instructional Science · doi:10.1007/bf00117714 · Cited by 4,285 · cited 1×
“Similarly, Sadler [33] described this distinction in terms of purpose and effect, noting that the purpose of formative assessment was to leverage judgements of student performance to enhance competency.”
Developing the theory of formative assessment
Black, P., & Wiliam, D. (2009) · Educational Assessment, Evaluation and Accountability · doi:10.1007/s11092-008-9068-5 · Cited by 3,301 · cited 2×
“From the aforementioned definition of formative assessment, peers, like teachers and learners, are also considered agents of the assessment, which implies that their decision is included in formative assessment [8].”
Formative assessment: a critical review
Bennett, R. E. (2011) · Assessment in Education: Principles, Policy & Practice · doi:10.1080/0969594x.2010.513678 · Cited by 1,498
Educational data mining and learning analytics: An updated survey
Romero, C., & Ventura, S. (2020) · WIREs Data Mining and Knowledge Discovery · doi:10.1002/widm.1355 · Cited by 1,120 · cited 2×
“Learning Analytics refers to the data-driven process of measuring, gathering, analyzing, and reporting on learners and educational contexts to better understand and enhance both learning and the environments where it takes place [22, 32].”
A Prompt Pattern Catalog to Enhance Prompt Engineering with ChatGPT
White, J., Fu, Q., Hays, S., Sandborn, M., Olea, C., Gilbert, H., Elnashar, A., Spencer-Smith, J., & Schmidt, D. C. (2023) · arXiv · doi:10.48550/arxiv.2302.11382 · Cited by 838
Perspectives of Curriculum Evaluation
Walbesser, H. H., Tyler, R. W., Gagné, R. M., Scriven, M., & Gagne, R. M. (1968) · American Educational Research Journal · doi:10.2307/1161709 · Cited by 792
Math anxiety in elementary and secondary school students
Wigfield, A., & Meece, J. L. (1988) · Journal of Educational Psychology · doi:10.1037/0022-0663.80.2.210 · Cited by 658
Research works printed but not linked: 3
- Jacqueline Bichsel. 2012. Analytics in higher education: Benefits, barriers, progress, and recommendations. EDUCAUSE Center for Applied Research.
- Murat Kahveci. 2010. Students’ Perceptions to Use Technology for Learning: Measurement Integrity of the Modified Fennema-Sherman Attitudes Scales. Turkish Online Journal of Educational Technology-TOJET 9 1 (2010) 185–201.
- Archie Zariski. 1996. Student peer assessment in tertiary education: Promise, perils and practice. In Teaching and Learning Within and Across Disciplines, p189-200. Proceedings of the 5th Annual Teaching and Learning Forum, Murdoch University.
Cited by
Xiao, N., Yuan, C. Z., Pei, Y., Wang, X., & Cai, Y. (2025) · Educational Studies · doi:10.1080/03055698.2025.2531969 · Cited by 50
“The personalisation of feedback represents another significant opportunity (Shin et al. 2025). found that students have distinct preferences regarding AI-generated versus peer-generated feedback in formative assessment contexts.”
Lin, X., & Tan, H. (2025) · Systems · doi:10.3390/systems13100840 · Cited by 22
Co-designing AI-powered learning analytics: bringing students and teachers together
Alfredo, R., Milesi, M. E., Echeverría, V., Gašević, D., Shum, S. B., Zhao, L., Yan, L., Jin, Y., Fan, J. X., Pammer‐Schindler, V., et al. (2025) · International Journal of Educational Technology in Higher Education · doi:10.1186/s41239-025-00572-8 · Cited by 18
“AI-powered features now enable personalized explanations of complex learning data (Yan et al., 2024a), automated feedback generation (Shin et al., 2025), and interactive dialogue about learning progress through chatbot interfaces (Jin et al., 2025).”
Özhan, Ş. Ç., Tekeli, P., & Kocadere, S. A. (2025) · Journal of Computer Assisted Learning · doi:10.1111/jcal.70134 · Cited by 13
Zhao, C. Q., Cao, J., Lin, J. X. C., & Koedinger, K. R. (2026) · International Conference on Learning Analytics and Knowledge (LAK) · doi:10.1145/3785022.3785124 · Cited by 7
“Along the human pathway, educator-authored feedback is rich and nuanced but constrained by time and workload, limiting scalability for targeted responses [1, 26]; peer or crowdsourced feedback can redistribute evaluative effort, yet often suffers from inconsistency and variable quality [7, 36].”
Metadata
- Authors
- Shin, I., Hwang, S. B., Yoo, Y. J., Bae, S., & Kim, R. Y.
- Affiliations
- Seoul National University (KR)
- Venue
- Proceedings of the 15th International Learning Analytics and Knowledge Conference
- Publisher
- ACM
- Published
- 2025-03-03
- Pages
- 159-169
- OpenAlex
- W4407841864
- Cited by
- 16
- References
- 43
- Citations by year
- 2026: 8 · 2025: 8
- FWCI
- 183.23
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- International Conference on Learning Analytics and Knowledge (LAK) (publishedVersion, cc-by)
- Topics
- Social Sciences › Education › Student Assessment and Feedback
- Social Sciences › Education › Artificial Intelligence in Education
- Physical Sciences › Computer Science › Artificial Intelligence › Intelligent Tutoring Systems and Adaptive Learning
- Keywords
- AI-generated feedback, peer assessment, formative assessment, student preferences, peer feedback, artificial intelligence in education, student perceptions
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-29, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-29 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Comparing Student Preferences for AI-Generated and Peer-Generated Feedback in AI-driven Formative Peer Assessment. https://app.cochat.ai/papers/10.1145/3706468.3706488
- The paper
- Shin, I., Hwang, S. B., Yoo, Y. J., Bae, S., & Kim, R. Y. (2025). Comparing Student Preferences for AI-Generated and Peer-Generated Feedback in AI-driven Formative Peer Assessment. Proceedings of the 15th International Learning Analytics and Knowledge Conference. https://doi.org/10.1145/3706468.3706488
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.