CoChat
Sign Up
Cited by 4Student Assessment and FeedbackArtificial Intelligence in EducationIntelligent Tutoring Systems and Adaptive Learning

Human-centered GenAI feedback design in higher education: a multisite experiment on direct, reflective, and hybrid approaches to scientific argumentation

Open access
Ates, Huseyin·2026·International Journal of Educational Technology in Higher Education
DOI 10.1186/s41239-026-00614-9
Open this paper in CoChat — free accountOpen PDF

Abstract

Generative artificial intelligence (GenAI) is increasingly used for feedback in higher education, yet evidence remains limited on how alternative human–AI feedback designs shape learning processes and durable outcomes. This study addresses that gap through a multisite, cluster-randomized, longitudinal field experiment comparing four feedback designs in introductory university science courses: peer feedback only, direct GenAI-supported feedback, reflective GenAI-supported feedback, and a hybrid design combining self-evaluation, peer feedback, and GenAI critique. The analytic sample comprised 1,176 first-year undergraduate students from 48 course sections across four universities and three science domains. Primary and secondary outcomes were argument-quality gain on four shared rubric dimensions—claim quality, evidence relevance and sufficiency, coherence of reasoning, and treatment of limitations or alternative explanations—conceptual learning, and delayed AI-free transfer; feedback uptake and self-regulated learning during revision were modeled as process mediators. Direct GenAI-supported feedback improved immediate argument-quality gain relative to peer feedback, whereas reflective and hybrid designs produced stronger feedback uptake and self-regulated learning. The hybrid condition yielded the highest adjusted mean for immediate argument-quality gain and showed the clearest advantage on conceptual learning; the reflective condition showed a positive but non-significant adjusted contrast on conceptual learning relative to direct GenAI-supported feedback. Both reflective and hybrid conditions outperformed direct GenAI-supported feedback on delayed AI-free transfer. Multilevel mediation analyses indicated that feedback uptake and self-regulated learning partially explained these advantages. By comparing four feedback designs, modeling revision processes, and assessing delayed AI-free transfer in a multisite field experiment, the findings suggest that the educational value of GenAI in higher education may depend less on AI access per se than on whether feedback environments preserve student agency, evaluative judgment, and ownership during revision.

AI snapshot

Written from the paper’s full text.

TL;DR

This multisite cluster-randomized field experiment compared peer-only, direct GenAI, reflective GenAI, and hybrid feedback designs for scientific argumentation in introductory university science courses. It found that direct GenAI improved immediate argument-quality gain over peer feedback, but reflective and hybrid designs produced stronger feedback uptake, self-regulated learning, conceptual learning, and delayed AI-free transfer.

Key findings

  • Among 1,176 first-year undergraduates in 48 course sections, immediate four-dimension argument-quality gain was highest in the hybrid condition (adjusted M = 4.31), followed by reflective GenAI (4.07), direct GenAI (3.41), and peer feedback only (2.84). Direct GenAI improved gain over peer feedback (b = 0.57, d = 0.25, p = .002), while reflective and hybrid both outperformed direct GenAI (reflective–direct b = 0.66, p < .001; hybrid–direct b = 0.90, p < .001); hybrid vs. reflective was not significant (p = .138).
  • Reflective and hybrid conditions yielded significantly higher feedback uptake and self-regulated learning than both peer feedback and direct GenAI, with no significant reflective–hybrid difference on either process variable. For example, feedback uptake contrasts: reflective–direct b = 0.26, p < .001; hybrid–direct b = 0.33, p < .001.
  • For conceptual learning, hybrid significantly outperformed direct GenAI (b = 0.25, p = .014), while reflective–direct was not significant after adjustment (p = .072). For delayed AI-free transfer, both reflective and hybrid significantly outperformed both peer feedback and direct GenAI, with no significant reflective–hybrid difference.
  • Multilevel mediation indicated that advantages of reflective and hybrid feedback over direct GenAI were partially explained by feedback uptake and self-regulated learning; for delayed AI-free transfer, all three indirect pathways—via uptake, via SRL, and via uptake → SRL—were statistically significant for both reflective and hybrid conditions.

Methods

  • Multisite, cluster-randomized, longitudinal field experiment across four universities and three science domains (biology, chemistry, physics). The unit of randomization was the course section; the analytic sample was 1,176 first-year undergraduates from 48 sections (12 sections per condition), drawn from an initial 1,248 students.
  • Four conditions: peer feedback only; direct GenAI-supported feedback; reflective GenAI-supported feedback (structured self-evaluation before AI critique); and hybrid feedback (self-evaluation, peer feedback, GenAI critique, and revision memo). The intervention ran 10 weeks with three argumentation cycles, using a common analytic rubric with four common content dimensions for gain and transfer plus a separate strategic-quality-of-revision indicator.
  • Main analyses used mixed-effects models with students at Level 1, sections as random intercepts, and institution as a fixed effect; outcomes included four-dimension argument-quality gain, conceptual learning, delayed AI-free transfer, feedback uptake, and SRL during revision. Mediation used 2-1-1 multilevel models; missing data were handled with multiple imputation by chained equations (20 imputations). Qualitative interviews with 32 students and 12 instructors were analyzed using reflexive thematic analysis.

Limitations

  • The authors state that reflective and hybrid conditions involved greater guided evaluative activity, so feedback-source configuration and reflective workload were not fully separable; AI-supported conditions also received additional system training, potentially increasing preparation time relative to peer feedback.
  • The authors note limited generalizability because the study was conducted in first-year university science courses; AI-free transfer cannot eliminate broader external familiarity with AI; the analytic sample may reflect consent-based selection; peer-comment quality was not independently scored; only four institutions were modeled as fixed effects; and moderation analyses were exploratory.
  • No additional limitations are made evident by the text beyond those stated by the authors.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1186/s41239-026-00614-9

Cites

22 of 22 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

42 of 53 cited sources checked · 19 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Self-assessment can positively influence self-regulated learning and self-efficacy because it requires learners to compare their work against explicit criteria and to generate internal judgments before receiving or acting on external feedback.

    “Meta-analytic findings indicate that self-assessment can positively influence self-regulated learning and self-efficacy because it requires learners to compare their work against explicit criteria and to generate internal judgments before receiving or acting on external feedback”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Panadero et al., 2017 source

      Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
  • Premisecitation

    Much of the literature on AI-supported feedback has focused on general writing tasks or broad comparisons among feedback sources, whereas disciplinary reasoning in university science courses has received less sustained attention.

    “much of the literature on AI-supported feedback has focused on general writing tasks or broad comparisons among feedback sources, whereas disciplinary reasoning in university science courses has received less sustained attention”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites Banihashem et al., 2024 source

      The Introduction frames the study as a comparison of peer- and AI-generated feedback for essay writing, and the Limitations section cautions against generalizing from its single course to other subjects. These passages support the characterization of this study, but they do not establish that much of the literature has that focus or that disciplinary reasoning in university science courses has received less attention. No page numbers are available in our copy. (LLM, conf=0.79)

      Experimental — accuracy not yet measured
    • cites Tang & Putra, 2025 source

      The Introduction discusses comparisons between customized and general-purpose chatbots, while the Research Site and Implementation section describes a study in Indonesian high schools. Neither addresses whether AI-feedback literature concentrates on general writing or comparisons among feedback sources, or whether disciplinary reasoning in university science courses has received less attention. Searches for feedback, writing, and university settings found no passage establishing that literature-wide claim. (LLM, conf=0.72)

      Experimental — accuracy not yet measured
    • cites Wang & Fan, 2025 source

      The Literature review (page unavailable) describes studies of ChatGPT-assisted text revision and writing, supporting the claim that writing tasks have received attention. It also describes research in STEM subjects and does not establish that feedback-source comparisons predominate or that disciplinary reasoning in university science courses has received less sustained attention. (LLM, conf=0.75)

      Experimental — accuracy not yet measured
  • Premisecitation

    Self-regulated learning is central in feedback contexts because students must decide how to interpret comments, whether to trust them, which revisions to prioritize, and how to integrate them into future performance.

    “SRL is central because students must decide how to interpret comments, whether to trust them, which revisions to prioritize, and how to integrate them into future performance.”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites Nicol & Macfarlane-Dick, 2006 source

      The conceptual model (p. 4) says students must interpret and internalise external feedback for it to affect subsequent learning. The discussion of closing the performance gap (p. 13) supports using feedback to plan improvements to later work. The copy does not specifically say students must decide whether to trust comments or which revisions to prioritise; its advice to prioritise areas for improvement is directed at feedback providers (p. 10). (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites Panadero, 2017 source

      The Winne and Hadwin model section describes students defining tasks, planning, choosing strategies, and adapting future learning through a feedback loop (page not provided). This supports the general role of self-regulation, but the review does not establish the specific claims about interpreting comments, deciding whether to trust them, or prioritizing revisions. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
  • Premisecitation

    Many studies compare a single AI condition with a single non-AI condition, leaving underexplored how alternative feedback designs may distribute evaluative work differently across learners, peers, and GenAI.

    “many studies compare a single AI condition with a single non-AI condition, leaving underexplored how alternative feedback designs may distribute evaluative work differently across learners, peers, and GenAI”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites Banihashem et al., 2024 source

      The Introduction describes this study as a comparison of peer-generated and ChatGPT-generated feedback, and the Discussion on RQ1 suggests that the two sources could play complementary roles. The paper therefore exemplifies a single AI-versus-non-AI comparison, but it does not establish that many studies use that design or specifically identify how alternative designs distribute evaluative work among learners, peers, and GenAI as an underexplored question. No page numbers are available in our copy. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites Chang et al., 2026 source

      The Introduction (page unavailable) says that how to use GenAI effectively for peer feedback remains underexplored, and the Research design section compares three feedback conditions. The paper does not say that many prior studies used only one AI and one non-AI condition, or frame the gap specifically as how evaluative work is distributed among learners, peers, and GenAI. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
    • cites Darvishi et al., 2024 source

      The Related Work section describes student agency in peer feedback as underexplored, and the study compares AI prompts, no prompts, self-monitoring checklists, and a hybrid condition. The Limitations section calls for further research on hybrid human–AI designs. Neither section establishes that many studies use only one AI and one non-AI condition, or specifically frames the gap as distributing evaluative work among learners, peers, and GenAI; this study’s AI prompts are not identified as GenAI. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
  • Premisecitation

    Immediate product improvement is often examined without simultaneously investigating feedback uptake, self-regulated learning, and later performance without AI support, making it difficult to determine whether observed gains reflect durable learning, stronger revision processes, or temporary performance support.

    “immediate product improvement is often examined without simultaneously investigating feedback uptake, self-regulated learning, and later performance without AI support, making it difficult to determine whether observed gains reflect durable learning, stronger revision processes, or temporary performance support”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites Wang & Fan, 2025 source

      Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites Xia et al., 2024 source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites Zhu et al., 2020 source
  • May combine two claims. 43
  • Classified as probably not a substantial claim. 47
  • The quote only partly states this claim. 4

  • May combine two claims.

    Premisecitation

    GenAI-supported feedback may help students improve written products and, when designed dialogically, support deeper reasoning, yet it may also reduce the need for students to make their own evaluative judgments if the feedback process encourages passive uptake.

    “GenAI-supported feedback is pedagogically promising but also potentially risky: it may help students improve written products and, when designed dialogically, support deeper reasoning, yet it may also reduce the need for students to make their own evaluative judgments if the feedback process encourages passive uptake”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites Darvishi et al., 2024 source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites Tang & Putra, 2025 source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites Zhu et al., 2020 source

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    GenAI can support learning performance, learning perceptions, and higher-order thinking, but these benefits vary substantially depending on the role assigned to the tool, the instructional design, and the characteristics of the task.

    “Existing evidence suggests that GenAI can support learning performance, learning perceptions, and higher-order thinking, but these benefits vary substantially depending on the role assigned to the tool, the instructional design, and the characteristics of the task”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Wang & Fan, 2025 source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    AI assistance in peer-feedback environments may increase dependence on the tool, raising questions about student agency in AI-supported learning contexts.

    “Darvishi et al. (2024) showed that AI assistance in peer-feedback environments may increase dependence on the tool, raising questions about student agency in AI-supported learning contexts.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Darvishi et al., 2024 source

  • Classified as probably not a substantial claim.

    Premisecitation

    Feedback becomes most productive when it stimulates internal feedback, that is, when learners engage in comparison processes between current performance, criteria, alternatives, and desired standards.

    “Nicol (2021) argued that feedback becomes most productive when it stimulates internal feedback, that is, when learners engage in comparison processes between current performance, criteria, alternatives, and desired standards.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Nicol, 2021 source

  • Classified as probably not a substantial claim.

    Premisecitation

    Feedback becomes productive when learners interpret feedback, compare it against criteria, judge its relevance, and use it to improve subsequent work.

    “it becomes productive when learners interpret feedback, compare it against criteria, judge its relevance, and use it to improve subsequent work”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites Carless & Boud, 2018 source
    • cites Hattie & Timperley, 2007 source
    • cites Nicol & Macfarlane-Dick, 2006 source

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Feedback becomes educationally meaningful when learners compare their current performance against criteria or standards, interpret the significance of comments, and use those judgments to guide subsequent action.

    “it becomes educationally meaningful when learners compare their current performance against criteria or standards, interpret the significance of comments, and use those judgments to guide subsequent action”
    Quote in the paperNot a substantial claim3 of 3 sources checked
    • cites Carless & Boud, 2018 source
    • cites Nicol & Macfarlane-Dick, 2006 source
    • cites Nicol, 2021 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The higher-education self-regulated learning tradition represented by the Motivated Strategies for Learning Questionnaire (MSLQ) and subsequent work emphasizes planning, monitoring, and strategic adaptation as core features of self-regulated learning.

    “the higher-education self-regulated learning tradition represented by the Motivated Strategies for Learning Questionnaire (MSLQ; Pintrich et al., 1991) and from subsequent work emphasizing planning, monitoring, and strategic adaptation as core features of SRL (Panadero, 2017).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites Pintrich et al., 1991

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites Panadero, 2017 source

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationcitation

    For higher education digital learning, the central implication is not whether students can access GenAI feedback, but how GenAI feedback is pedagogically organized so that human evaluative work remains at the center.

    “For higher education digital learning, the central implication is therefore not whether students can access GenAI feedback, but how GenAI feedback is pedagogically organized so that human evaluative work remains at the center (Buckingham Shum et al., 2023; Xia et al., 2024).”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites Buckingham Shum et al., 2023 source

      The Introduction and Conclusion emphasize designing automated feedback around active student participation and teachers’ skilled use of technology; the ‘Crafts appropriate inputs to students’ section says such tools should augment teachers rather than automate them out of the loop. This supports the claim’s focus on pedagogical organization and human work, but the paper does not frame that implication specifically around students’ access to GenAI feedback. (LLM, conf=0.85)

      Experimental — accuracy not yet measured
    • cites Xia et al., 2024 source

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Recent work has suggested that the educational use of automated feedback depends on how human and machine roles are distributed within feedback processes, and that AI assistance may weaken student agency when critique is accepted too readily or when difficult judgments are outsourced to the system.

    “Recent work has suggested that the educational use of automated feedback depends on how human and machine roles are distributed within feedback processes, and that AI assistance may weaken student agency when critique is accepted too readily or when difficult judgments are outsourced to the system (Buckingham Shum et al., 2023;Darvishi et al., 2024).”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites Buckingham Shum et al., 2023 source

      Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured
    • cites Darvishi et al., 2024 source

      Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationcitation

    The study adds nuance to recent syntheses of GenAI in higher education by suggesting that apparent benefits are highly design-sensitive and that concerns about authorship, evidence of learning, and learner responsibility are not only matters of policy or integrity, but also matters of feedback design.

    “More broadly, the study adds nuance to recent syntheses of GenAI in higher education by suggesting that apparent benefits are highly design-sensitive and that concerns about authorship, evidence of learning, and learner responsibility are not only matters of policy or integrity, but also matters of feedback design(Wang & Fan, 2025;Xia et al., 2024).”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites Wang & Fan, 2025 source
    • cites Xia et al., 2024 source

      Split into 2 atoms — 1 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    The Student Feedback Literacy Instrument (SFLI) was specifically developed for use with higher-education students and builds on earlier initial scale-validation work conducted in the same context.

    “The instrument was selected because it was specifically developed for use with higher-education students and builds on earlier initial scale-validation work conducted in the same context (Woitt et al., 2025).”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites Weidlich et al., 2025 source
    • cites Woitt et al., 2025 source

  • The quote only partly states this claim.

    Findingown data

    Hybrid feedback produced larger four-dimension argument-quality gains than peer feedback.

    “both reflective and hybrid feedback produced still larger gains.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Missing data are missing at random.

  • May combine two claims.

    Findingown data

    In the present study, direct GenAI-supported feedback supported immediate improvement in four-dimension argument-quality gain, but reflective and hybrid designs were more consistently associated with stronger feedback uptake, self-regulated learning during revision, conceptual learning, and AI-independent transfer.

    “In the present study, direct GenAI-supported feedback supported immediate improvement in four-dimension argument-quality gain, but reflective and hybrid designs were more consistently associated with stronger feedback uptake, self-regulated learning during revision, conceptual learning, and AI-independent transfer.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The more agentic designs were associated not only with larger gains in common-content argument quality, but also with stronger supplementary indicators of revision substance and revised-draft strategic quality.

    “The distinction between four-dimension argument-quality gain and the strategic quality of revision sharpens this interpretation further: the more agentic designs were associated not only with larger gains in common-content argument quality, but also with stronger supplementary indicators of revision substance and revised-draft”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    Hybrid feedback significantly outperformed peer feedback on delayed AI-free transfer.

    “whereas both reflective and hybrid feedback significantly outperformed both peer feedback and direct GenAI-supported feedback.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Missing data are missing at random.

  • May combine two claims.

    Findingown data

    Participants often noted that self-, peer-, and AI-based feedback sources highlighted different weaknesses in the same draft, thereby supporting more deliberate judgment about which revisions to prioritize.

    “Participants often noted that these sources highlighted different weaknesses in the same draft, thereby supporting more deliberate judgment about which revisions to prioritize.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    In the present study, direct GenAI-supported feedback provided rapid, criterion-referenced critique, but it did so without requiring learners to externalize an initial judgment about the quality of their own work.

    “In the present study, direct GenAI-supported feedback provided rapid, criterion-referenced critique, but it did so without requiring learners to externalize an initial judgment about the quality of their own work.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The more agentic conditions were also associated with more substantive revision and stronger revised-draft strategic quality of revision.

    “Although the primary immediate outcome was defined as four-dimension argument-quality gain, the more agentic conditions were also associated with more substantive revision and stronger revised-draft strategic quality of revision.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Argument-quality gain was highest in the hybrid condition, followed by the reflective GenAI-supported feedback condition, the direct GenAI-supported feedback condition, and the peer-feedback condition.

    “argument-quality gain was highest in the hybrid condition, followed by the reflective GenAI-supported feedback condition, the direct GenAI-supported feedback condition, and the peerfeedback condition.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Missing data are missing at random.

  • May combine two claims.

    Findingown data

    Students described a sense that AI-enabled feedback reduced the effort required to generate revision ideas while sometimes weakening their sense of authorship over the final text.

    “The fourth theme, tensions between efficiency and ownership of revision, described students' sense that AIenabled feedback reduced the effort required to generate revision ideas while sometimes weakening their sense of authorship over the final text.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    Reflective GenAI-supported feedback significantly outperformed direct GenAI-supported feedback on delayed AI-free transfer.

    “whereas both reflective and hybrid feedback significantly outperformed both peer feedback and direct GenAI-supported feedback.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Missing data are missing at random.

  • May combine two claims.

    Findingown data

    The present results are consistent with those concerns: students in the direct GenAI-supported feedback condition described AI as fast and useful, but sometimes overly directive, whereas students in the reflective and hybrid conditions described self-evaluation, peer input, and revision justification as preserving a stronger sense of ownership.

    “The present results are consistent with those concerns: students in the direct GenAI-supported feedback condition described AI as fast and useful, but sometimes overly directive, whereas students in the reflective and hybrid conditions described self-evaluation, peer input, and revision justification as preserving a stronger sense of ownership.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    Reflective GenAI-supported feedback significantly outperformed peer feedback on delayed AI-free transfer.

    “whereas both reflective and hybrid feedback significantly outperformed both peer feedback and direct GenAI-supported feedback.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Missing data are missing at random.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Chat history was reset between cycles to avoid cross-task carryover.

    “Chat history was reset between cycles to avoid cross-task carryover.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Scientific argumentation requires students to formulate claims, justify them with evidence and reasoning, consider alternative explanations, and revise disciplinary explanations in response to critique.

    “Scientific argumentation requires students not only to produce acceptable answers, but also to formulate claims, justify them with evidence and reasoning, consider alternative explanations, and revise disciplinary explanations in response to critique”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Driver et al., 2000

      Not checked in the free run.

    • cites Osborne et al., 2004 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    No inter-coder reliability coefficient was calculated, in keeping with the reflexive thematic analysis approach.

    “No inter-coder reliability coefficient was calculated, in keeping with the reflexive thematic analysis approach.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    Implementation fidelity was monitored through four procedures: common intervention schedule, rubric, and instructional materials across sites; standardized prompt architecture and fixed system settings in AI-supported conditions; platform logs verifying access only to assigned feedback resources; and manual audits of a random 15% sample of student cases across sites.

    “Implementation fidelity was monitored through four procedures: use of a common intervention schedule, rubric, and instructional materials across sites; standardized prompt architecture and fixed system settings in the AI-supported conditions; platform logs verifying that students accessed only the feedback resources assigned to their condition; and manual audits of a random 15% sample of student c”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Instructors were not permitted to alter prompts during the intervention period.

    “Instructors were not permitted to alter prompts during the intervention period.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Allocation was completed after course enrollment closed and before the Week 1 pre-test.

    “Allocation was completed after course enrollment closed and before the Week 1 pre-test.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductnone given

    Tasks were co-developed by participating instructors through a shared design protocol and reviewed in cross-institution calibration meetings focusing on epistemic demands, expected difficulty, and rubric fit across biology, chemistry, and physics.

    “Tasks were co-developed by participating instructors through a shared design protocol and reviewed in cross-institution calibration meetings focusing on epistemic demands, expected difficulty, and rubric fit across biology, chemistry, and physics.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductown data

    Sections were randomly assigned to one of the four feedback conditions, resulting in 12 sections per condition.

    “Sections were randomly assigned to one of the four feedback conditions, resulting in 12 sections per condition.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    In introductory university science courses, students are expected to formulate claims, coordinate claims with evidence and reasoning, address limitations or counterpositions, and revise explanations in light of critique.

    “In introductory university science courses, students must do more than produce coherent text.They are expected to formulate claims, coordinate claims with evidence and reasoning, address limitations or counterpositions, and revise explanations in light of critique”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Driver et al., 2000

      Not checked in the free run.

    • cites Osborne et al., 2004 source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductreasoning

    To preserve balance, randomization was blocked by institution and discipline, and section meeting time was used as a secondary balancing variable.

    “To preserve balance, randomization was blocked by institution and discipline, and section meeting time (morning/afternoon) was used as a secondary balancing variable.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Access to condition-specific feedback resources was monitored through platform records, and potential contamination through non-assigned resources is addressed under implementation fidelity.

    “Access to condition-specific feedback resources was monitored through platform records, and potential contamination through non-assigned resources is addressed under implementation fidelity.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    All AI-supported conditions used the same prompt architecture, rubric anchors, interface settings, and fixed model parameters across all three intervention cycles to ensure consistency across institutions.

    “To ensure consistency across institutions, all AIsupported conditions used the same prompt architecture, rubric anchors, interface settings, and fixed model parameters across all three intervention cycles.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning

    The unit of randomization was the course section, whereas the primary unit of analysis was the student, with clustering handled through multilevel modeling.

    “The unit of randomization was the course section, whereas the primary unit of analysis was the student, with clustering handled through multilevel modeling.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    To assess robustness, the primary models were re-estimated using per-protocol analyses, models excluding low-fidelity sections, and an alternative missing-data specification based on full-information maximum likelihood.

    “To assess robustness, the primary models were also re-estimated using per-protocol analyses, models excluding low-fidelity sections, and an alternative missing-data specification based on full-information maximum likelihood.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Missing data were handled using multiple imputation by chained equations with 20 imputations under a missing-at-random assumption.

    “Missing data were handled using multiple imputation by chained equations with 20 imputations under a missing-at-random assumption.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Missing data are missing at random.

  • May combine two claims.

    Study design and conductown data

    Audits were conducted by trained research assistants not involved in instruction, and 20% of audited cases were double-coded, with high agreement on checklist items (Cohen's κ = .89).

    “Audits were conducted by trained research assistants who were not involved in instruction, and 20% of audited cases were double-coded, with high agreement on checklist items (Cohen's κ = .89).”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Fidelity was assessed using a predefined checklist covering sequence integrity, timing compliance, access restrictions, and completion of condition-specific activities.

    “Fidelity was assessed using a predefined checklist covering sequence integrity, timing compliance, access restrictions, and completion of condition-specific activities.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    The qualitative component was interpretive rather than reliability-seeking.

    “The qualitative component was interpretive rather than reliability-seeking.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    The design, primary outcomes, focal contrasts, and analysis plan were preregistered on the Open Science Framework (OSF) prior to post-test analysis.

    “The design, primary outcomes, focal contrasts, and analysis plan were preregistered on the Open Science Framework (OSF) prior to post-test analysis.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    Because research consent was voluntary while instructional participation was compulsory, the analytic sample may reflect a degree of consent-based selection.

    “Because research consent was voluntary while instructional participation was compulsory, the analytic sample may reflect a degree of consent-based selection.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    All primary analyses followed an intention-to-treat approach.

    “All primary analyses followed an intention-to-treat approach.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductreasoning

    Students were limited to a single feedback generation per draft and could not iteratively regenerate responses within the same cycle, although they could review the generated feedback during revision.

    “To reduce variability and prompt gaming, students were limited to a single feedback generation per draft and could not iteratively regenerate responses within the same cycle, although they could review the generated feedback during revision.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    To reduce institution-level heterogeneity, the study used a common intervention schedule, shared task-design procedures, and a common analytic rubric across sites.

    “To reduce institution-level heterogeneity, the study used a common intervention schedule, shared task-design procedures, and a common analytic rubric across sites.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductnone given

    Prior to data collection, raters completed two calibration sessions using anchor responses from all three disciplinary domains.

    “Prior to data collection, raters completed two calibration sessions using anchor responses from all three disciplinary domains.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductnone given

    Randomization was conducted by a researcher not involved in instruction using a computer-generated random-number sequence.

    “Randomization was conducted by a researcher not involved in instruction using a computergenerated random-number sequence.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    Internal consistency in the present sample for the SFLI was high (McDonald's ω = .90).

    “Internal consistency in the present sample was high (McDonald's ω = .90).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Study design and conductnone given

    Novelty of the delayed transfer task was ensured by using prompts that had not appeared in course activities, practice materials, or post-test instruments.

    “Novelty was ensured by using prompts that had not appeared in course activities, practice materials, or post-test instruments.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductnone given

    Two trained raters independently scored all first and final drafts used in outcome assessment, and discrepancies greater than one point on any rubric dimension were resolved through discussion.

    “Two trained raters independently scored all first and final drafts used in outcome assessment, and discrepancies greater than one point on any rubric dimension were resolved through discussion.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    The advantages of reflective and hybrid feedback were not limited to larger gain scores alone; they were also linked to revision that appeared more purposeful, better integrated, and more epistemically substantive.

    “This pattern suggests that the advantages of reflective and hybrid feedback were not limited to larger gain scores alone; they were also linked to revision that appeared more purposeful, better integrated, and more epistemically substantive.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    At the institutional level, the study suggests that decisions about GenAI adoption should not be made only in terms of scalability or efficiency, but also in terms of whether digital-learning designs preserve student agency, evaluative judgment, and ownership of learning when AI tools are embedded in routine coursework.

    “At the institutional level, the study suggests that decisions about GenAI adoption should not be made only in terms of scalability or efficiency, but also in terms of whether digital-learning designs preserve student agency, evaluative judgment, and ownership of learning when AI tools are embedded in routine coursework.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductreasoning

    Although peer feedback was structured through a common template, peer-comment quality was not independently scored as a separate explanatory variable, making it difficult to determine the extent to which condition differences may partly reflect variation in the quality of peer input.

    “Sixth, although peer feedback was structured through a common template, peer-comment quality was not independently scored as a separate explanatory variable, making it difficult to determine the extent to which condition differences may partly reflect variation in the quality of peer input.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    The participating institutions were modeled as fixed effects because only four institutions were included.

    “Seventh, the participating institutions were modeled as fixed effects because only four institutions were included.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    The internal consistency of the self-report measures was satisfactory, and inter-rater reliability for scientific argumentation scoring and revision-depth coding was high.

    “The internal consistency of the self-report measures was satisfactory, and inter-rater reliability for scientific argumentation scoring and revision-depth coding was high.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductnone given

    The study was conducted in first-year university science courses, and the extent to which the findings generalize to other disciplines, levels of study, or institutional contexts remains uncertain.

    “Third, the study was conducted in first-year university science courses, and the extent to which the findings generalize to other disciplines, levels of study, or institutional contexts remains uncertain.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Taken together, these findings support the adequacy of the dataset and the integrity of the intervention.

    “Taken together, these findings support the adequacy of the dataset and the integrity of the intervention.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Cross-condition contamination was minimal.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Hybrid feedback designs appear particularly well suited to complex reasoning tasks in which students must compare perspectives, weigh evidence, and justify revision decisions.

    “Second, hybrid feedback designs appear particularly well suited to complex reasoning tasks in which students must compare perspectives, weigh evidence, and justify revision decisions.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    Baseline equivalence analyses indicated that the four conditions did not differ significantly on age, gender distribution, prior knowledge, baseline scientific argumentation, prior achievement, feedback literacy, or prior GenAI experience (all p ≥ .472), suggesting that the randomization procedure produced broadly comparable groups prior to the intervention.

    “Baseline equivalence analyses indicated that the four conditions did not differ significantly on age, gender distribution, prior knowledge, baseline scientific argumentation, prior achievement, feedback literacy, or prior GenAI experience (all p ≥ .472),suggesting that the randomization procedure produced broadly comparable groups prior to the intervention.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    Final sample sizes were balanced across conditions, and attrition did not differ significantly by condition, χ2(3) = 0.33, p = .954.

    “Final sample sizes were balanced across conditions, and attrition did not differ significantly by condition, χ²(3) = 0.33, p = .954.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    The pattern of no statistically significant differences between reflective and hybrid conditions on several outcomes is consistent with the possibility that requiring learners to externalize an initial evaluative judgment before receiving AI critique is itself a particularly important ingredient of effective GenAI-supported feedback.

    “Rather than demonstrating equivalence in any strong sense, the pattern is consistent with the possibility that requiring learners to externalize an initial evaluative judgment before receiving AI critique is itself a particularly important ingredient of effective GenAI-supported feedback.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The more agentic feedback designs improved learning outcomes not only by changing the feedback resources available to students, but also by changing how students engaged with and acted on those resources.

    “Taken together, these findings indicate that the more agentic feedback designs improved learning outcomes not only by changing the feedback resources available to students, but also by changing how students engaged with and acted on those resources.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Missing data are missing at random.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    The moderation analyses were exploratory and should therefore be interpreted with caution.

    “Finally, the moderation analyses were exploratory and should therefore be interpreted with caution.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Study design and conductnone given

    AI-free task completion meant that students had no access to the study's GenAI tool and completed the task independently in a supervised classroom setting without internet-enabled devices.

    “In the present study, AI-free task completion meant that students had no access to the study's GenAI tool and completed the task independently in a supervised classroom setting without internet-enabled devices.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning

    The reflective and hybrid conditions involved greater amounts of guided evaluative activity, meaning that feedback-source configuration and reflective workload were not fully separable.

    “First, the reflective and hybrid conditions involved greater amounts of guided evaluative activity, meaning that feedback-source configuration and reflective workload were not fully separable.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Premisecitation

    Reflexive thematic analysis following Braun and Clarke (2006) is an appropriate approach for interpretive qualitative analysis.

    “analyzed using reflexive thematic analysis following Braun and Clarke (2006).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Braun and Clarke (2006) source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    Students in AI-supported conditions received additional system training, which may have increased condition-specific preparation time relative to the peer-feedback condition.

    “Second, students in AI-supported conditions received additional system training, which may have increased condition-specific preparation time relative to the peerfeedback condition.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    Of the 1,248 students initially enrolled, 72 were excluded because of non-consent for research use of their data, course withdrawal before the first intervention cycle, or absence from both the post-test and delayed transfer sessions.

    “Of the 1,248 students initially enrolled, 72 were excluded because of non-consent for research use of their data, course withdrawal before the first intervention cycle, or absence from both the post-test and delayed transfer sessions.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductown data

    Missing data on the retained sample ranged from 0.6% to 3.4% across variables and were handled using multiple imputation with 20 imputations.

    “Missing data on the retained sample ranged from 0.6% to 3.4% across variables and were handled using multiple imputation with 20 imputations.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    When the goal is quick draft improvement, direct GenAI-supported feedback may be useful because it provides rapid, criterion-referenced critique with relatively low orchestration demands.

    “When the goal is quick draft improvement, direct GenAI-supported feedback may be useful because it provides rapid, criterion-referenced critique with relatively low orchestration demands.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The results imply that GenAI feedback systems should be designed around comparison, selection, and justification rather than passive implementation of suggestions.

    “More broadly, the results imply that GenAI feedback systems should be designed around comparison, selection, and justification rather than passive implementation of suggestions.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationown data

    The study contributes to emerging discussions of epistemic agency in AI-supported learning.

    “The study also contributes to emerging discussions of epistemic agency in AI-supported learning.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductreasoning

    Although AI-free transfer tasks were completed without access to the study's GenAI system in supervised settings, the design cannot eliminate broader concerns about students' external familiarity with AI beyond the intervention context.

    “Fourth, although AI-free transfer tasks were completed without access to the study's GenAI system in supervised settings, the design cannot eliminate broader concerns about students' external familiarity with AI beyond the intervention context.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    GenAI can support rapid improvement in common-content argument quality, but more agentic feedback designs are more effective when the goal is durable learning and AI-independent transfer.

    “Taken together, these findings suggest that GenAI can support rapid improvement in common-content argument quality, but more agentic feedback designs are more effective when the goal is durable learning and AI-independent transfer.”
    Quote in the paperSubstantial claim
75 of 75

  • The paper relies on this without support.

    The GenAI tool provides criterion-referenced feedback of adequate quality to support the intended learning processes.

    “The system generated criterion-referenced feedback on claim quality, evidence use, reasoning, limitations, and possible revision moves.”

  • The paper relies on this without support.

    The delayed AI-free transfer task measures durable learning rather than temporary performance.

    “Secondary outcomes were conceptual learning and delayed AI-free transfer.”

  • The paper relies on this without support.

    Feedback uptake and self-regulated learning during revision are valid mediators of the relationship between feedback design and learning outcomes.

    “Two process variables-feedback uptake and self-regulated learning during revision-were specified a priori as mediators linking feedback design to learning outcomes.”

  • The paper relies on this without support.

    The four feedback conditions were implemented with fidelity across all sites and sections.

    “An overview of the four intervention conditions is provided in Table 1.”

  • The authors state this as an assumption.

    Missing data are missing at random.

    “under a missing-at-random assumption.”

  • The paper relies on this without support.

    Cross-condition contamination was minimal.

    “Cross-condition contamination was minimal.”
6 of 6

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • ConsistentF(3, 43.8) = 18.76reported p < .001 · recomputed p < .001

    “F(3, 43.8) = 18.76, p < .001”

  • ConsistentF(3, 43.5) = 7.62reported p < .001 · recomputed p < .001

    “F(3, 43.5) = 7.62, p < .001”

  • ConsistentF(3, 43.9) = 12.03reported p < .001 · recomputed p < .001

    “F(3, 43.9) = 12.03, p < .001”

  • ConsistentF(3, 43.6) = 16.28reported p < .001 · recomputed p < .001

    “F(3, 43.6) = 16.28,p < .001”

  • ConsistentF(3, 43.2) = 11.91reported p < .001 · recomputed p < .001

    “F(3, 43.2) = 11.91,p < .001”

  • ConsistentF(3, 1158) = 2.83reported p = .037 · recomputed p = .037

    “F(3, 1158) = 2.83, p = .037”

  • Consistentχ²(3) = 0.33reported p = .954 · recomputed p = .954

    “χ2(3) = 0.33, p = .954”

7 of 7

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.

  • Baseline p-value summary contradicts Table 3

    Minor

    The paper says all baseline-equivalence p values were at least .472, but Table 3 reports p = .236 for prior GenAI experience and p = .349 for baseline scientific argumentation.

1 of 1

Cited works that were retracted

Works this paper cites that were later retracted, from the retraction notices publishers deposit. Major when the notice came out before this paper was published.

  • Cites a retracted work: 10.1057/s41599-025-04787-y

    Major

    The cited work was retracted on 2026-04-22 (notice 10.1057/s41599-026-07310-z), before this paper was published (2026-07-16). Its authors could have known.

    Cited work 10.1057/s41599-025-04787-y · retracted 2026-04-22 · notice 10.1057/s41599-026-07310-z · this paper published 2026-07-16

1 of 1

Cites

1 retracted

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Ates, H.
Venue
International Journal of Educational Technology in Higher Education
Publisher
Springer Science and Business Media LLC
Published
2026-07-16
Volume
23
Issue
1
ISSN
2365-9440 (ISSN-L)
DOI
10.1186/s41239-026-00614-9
OpenAlex
W7168352747
Cited by
4
References
21
Citations by year
2026: 4
FWCI
84.00
Type
journal article
Language
en
Open access
Licence
Locations
  • International Journal of Educational Technology in Higher Education (publishedVersion, cc-by)
  • Research Square (acceptedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
Topics
  • Social Sciences › Education › Student Assessment and Feedback
  • Physical Sciences › Computer Science › Information Systems › Artificial Intelligence in Education
  • Physical Sciences › Computer Science › Artificial Intelligence › Intelligent Tutoring Systems and Adaptive Learning
Keywords
generative AI, AI-generated feedback, scientific argumentation, peer feedback, reflective feedback, self-assessment, feedback uptake, student self-regulation, conceptual learning, argument quality, student agency, revision

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.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-11-02.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-11-02.

Cited sources are checked by gpt-6-sol.

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