Concept
Feedback
Feedback — information provided to a learner about their performance or understanding that is intended to close the gap between current and desired performance. In AI in education, feedback has become a central and rapidly transforming theme: AI systems now generate, deliver, and even teach students how to use feedback, reshaping every stage of the feedback process.
Questions to Consider
- Feedback is often imagined as information handed to a learner, but research argues it only 'counts' as feedback when the learner makes sense of and acts on it. If feedback is a relationship rather than a message, what changes about how we should deliver it?
- Students' perception of whether feedback came from an AI or a human can change how much they learn from it. Why do you think the source matters so much — and what does that imply for AI-generated feedback in your context?
- Immediate feedback can be a trap: the 'correct-answer trap' shows that students who simply copy corrections learn less. When does instant feedback help, and when does it short-circuit the very learning it's meant to support?
- Research even found that a well-designed sequence of feedback — encouragement, then hints, then the answer — designed to build autonomy actually harmed learning despite boosting engagement. If students feel good about feedback that makes them learn less, what should we optimize?
- Feedback is described as a system with parts: the feedback loop, quality of provision, the learner's uptake capacity, and the assessment context. If you were trying to improve feedback in your own course, which part would you fix first, and why?
Introduction
Feedback is the umbrella concept for a cluster of related ideas. Feedback sits at the intersection of assessment and learning: without feedback, assessment measures performance but does not improve it; with effective feedback, assessment becomes a learning event. The research treats feedback as a system with multiple facets — the quality of the feedback itself (AI feedback quality), the loop through which it closes the learning gap, the learner's capacity to use it (feedback literacy), and the assessment contexts in which it operates (formative assessment, peer assessment, automated assessment).
- Yilmaz et al. show that students' perception of the feedback source (AI vs human) shapes self-regulated learning from GenAI feedback — a crucial qualifier for feedback effectiveness claims.
- Nazaretsky et al. (2025) isolate the source from the content: 472 EPFL students rated content-comparable human and GPT-4.0 feedback on their own work, first blind and then again after the source was disclosed, and most could not reliably tell the two apart (287 of 472 guessed correctly). Once the label was revealed, ratings of the AI variant fell and of the human variant rose — significantly for Genuineness (p < .01) — and provider credibility separated cleanly (human μ = 3.28 vs. AI μ = 2.25, Cohen's d = 0.57). The bias attaches to the provider rather than the text, since the two variants were comparable on correctness and cognitive quality. Pérez-López et al. (2026) carry feedback credibility through to learning: among 230 business students after an AI-supported essay, credibility predicted perceived learning, with perceived competence and human-like empathy as its antecedents.
The feedback system
Feedback is best understood not as a single event but as a connected system of interacting parts, each of which carries its own evidence:
- The feedback loop — the mechanism through which feedback closes the gap between current and desired performance; the cycle of performance → feedback → revision → improved performance that drives learning.
- AI feedback quality — the accuracy, usefulness, timeliness, and pedagogical value of feedback generated by AI; the provision side of the system.
- feedback literacy — the capabilities and dispositions students need to understand, evaluate, and act on feedback; the uptake side of the system.
- formative assessment — the assessment context in which feedback is used to improve learning while it is still in progress, rather than merely to judge it.
- peer assessment — feedback exchanged between learners, increasingly augmented by AI and a site for developing feedback literacy.
- automated assessment — AI-driven scoring and feedback, from automated essay scoring to confidence-aware short-answer grading.
- The parts trade off against each other. Designing feedback means choosing among usability, sources and sustainability rather than maximizing all three; Pedlow et al. (2026) treat sustainability as the pole that must not depend on any single individual, resource or institutional condition.
The feedback loop
The feedback loop is the cyclical process where AI systems assess student work, deliver feedback, observe the student's response, and adapt subsequent instruction. AI-mediated feedback loops operate at multiple timescales:
- Immediate feedback: Automated grading systems and AI tutors provide real-time correction during problem-solving. Correct-answer trap research shows that immediate feedback can short-circuit learning if students simply copy corrections. Generating that immediate corrective feedback with LLMs inside a tutor is feasible but imperfect: across 6,926 logged transactions in the Apprentice Tutor College Algebra platform, Reddig, Arora & MacLellan (2025) found GPT-4 produced hints targeted to the student's specific error ~66% of the time, yet ~35% of hints were too general, incorrect, or prematurely gave away the answer, and LLM-based automated quality checks misaligned with human judgment — reinforcing that unvetted AI feedback in the immediate loop carries real short-circuiting risk.
- Assignment-level feedback: formative assessment systems and AI feedback quality research examine whether AI-generated assignment feedback improves subsequent work. Sequenced feedback studies test whether the order of feedback matters.
- Course-level loops: Learning analytics dashboards and educational platforms aggregate feedback across assignments to identify patterns and recommend interventions.
The effectiveness of a feedback loop depends on feedback quality — accuracy, specificity, timeliness, and actionability. AI peer feedback systems add a social dimension to the loop.
The human in the loop: feedback loops are not purely automated — teachers often mediate AI-generated feedback before it reaches learners. Studies of AI feedback tools for teachers (e.g., the PolyFeed tool combining an ML detector with an LLM rephraser) find teachers use professional judgment to accept, edit, or reject AI suggestions — an "assist but verify" pattern — and systematically moderate exaggerated praise and generic suggestions to protect authenticity and voice. The relational/affective dimension of feedback (student–teacher relationship, encouragement) most strongly resists AI delegation, suggesting this part of the loop remains inherently human. This human-in-the-loop mediation connects to human-in-the-loop and to teacher role.
- The relational penalty is measurable, and separable from quality. In a three-arm RCT with 287 writers, AI and instructor feedback raised revised-essay quality equally (d = 0.60 and 0.59), but instructor feedback raised intrinsic motivation through perceived relatedness (Liu & Shi (2026)).
Luo and Eaton's policy review sharpens the same point: only 14 of 50 top-ranked universities had specific guidance on teachers' AI use for feedback, and students described undisclosed AI feedback as "disingenuous", making transparency and consent preconditions of ethical practice rather than courtesies (Luo & Eaton (2026)).
LearnLens turns that mediation into a mechanism: it treats a teacher's edit as a signal of dissatisfaction rather than a correction, spending additional compute on deeper reflection instead of returning the fast answer, and its teachers' median time per assignment fell from 10–30 minutes to under five (Zhao et al. (2025)).
The relational reading cuts deeper than substitution. In workshops with 12 students and 18 educators, Bearman et al. (2026) found students treated AI as an additional but flawed source, weighing it against rubrics, lecture notes, discussion boards and educator comments, and using it as a holding pattern when human responses were slow; one pasted thin educator comments into a chatbot to work out what to do next. AI's presence changed relations rather than merely adding comments, to the point that one student stopped trusting peer discussion-board feedback as probably AI-written. The authors therefore propose humans-as-the-loop: people helping people build stronger feedback relationships over time, in place of oversight of machine output.
Winstone et al. (2026) reframe GenAI feedback as matters of care — an ethical, relational practice rather than information transfer — and name a safe-space paradox: the tool's reduced personal risk may forgo priming students for the friction of professional feedback encounters.
How AI transforms feedback
AI changes feedback in three consequential directions, each raising the stakes of the other facets:
-
Volume and immediacy. AI can generate feedback instantly and at scale, dramatically increasing how much feedback students receive. Multi-site GenAI feedback studies and AI feedback in higher education document AI feedback experienced as comparable to teacher feedback in acceptability and supportiveness. A PRISMA-guided systematic review of 42 empirical studies (2023–2025) corroborates this scale benefit — LLMs can accelerate grading and deliver rapid, personalized feedback at scale, especially in large or higher-education cohorts — while cautioning that such feedback is sometimes too generic or misaligned with the assigned grade, and that reliability slips on longer, multilingual, or nuanced tasks (Jukiewicz et al., 2026).
-
Personalization is a bias vector. Holding eighth-grade essays constant, four LLMs over-praised and withheld critique for students marked by race, language, or disability — so which student attributes feed a feedback system is a fairness decision, not a neutral one (Tan, Phalen & Demszky (2026)).
-
Evaluation demands on the learner. Because AI feedback can be inaccurate or hallucinated, students must judge whether to trust and act on it. This is where feedback literacy and trust calibration become decisive: Mendoza et al. (2026) show that only feedback-literate students convert AI feedback into self-regulated learning gains, while the LLM fallacy captures how students may over-credit AI feedback to their own competence.
-
Feedback literacy is built by use, not demographics. In 486 EFL undergraduates, only frequency of AI tool use predicted AI feedback literacy (β = 0.29) — not major or year of study — so scaffolding access to AI-mediated feedback is an equity question as much as a design one (Liu & Deris (2025)).
-
Lab approval is not semester use. All 18 students reacted positively to the GenAI explanation function in lab sessions, yet only half used it during the term, citing mismatch, irrelevance, and redundancy when educator feedback was already clear (Jin et al. (2025)).
-
Trust and care, not accuracy, bound uptake. Seventeen EFL students granted an AI writing assistant high expertise yet doubted its trustworthiness over data handling and rated its goodwill lowest, authorizing it only for broad language concerns; several also found its volume of suggestions unmanageable (Serpil & Mor, 2026).
-
Task-level suitability: sorting feedback by epistemic status. Venetsanos (2026) argues that the frameworks the field relies on — Hattie and Timperley's levels, Boud and Molloy's design principles — explain where feedback operates in learning but not whether a given feedback task involves rule-checking, factual verification, or interpretive judgment, and that it is the second distinction that decides whether automation is defensible. The proposed tripartite split is low-level structural and presentational feedback (formatting, referencing, grammar, sentence-level clarity — rule-based against explicit standards and safely automatable), intermediate-level factual and procedural verification (whether a calculation used a specified method correctly, whether a stated fact matches authoritative references — comparison against established knowledge, not judgment of reasoning), and high-level critical evaluation and synthesis, where judgment is interpretive and must stay human. The levels are dimensions rather than a ladder: one calculation can be correctly presented, arithmetically wrong, and methodologically inappropriate at the same time, and can attract verification, process feedback and self-regulation feedback together. The framework's sharpest boundary case is clarity: sentence-level clarity is surface-level, but argumentative unclarity often signals incomplete conceptual grasp, so flagged instances should escalate to human review rather than being corrected as prose — fixing the writing would leave the actual problem untouched.
-
Role-aware feedback. Yaşar et al. (2026) showed that prompting the same LLM to evaluate the same student artifact under different roles — instructor, peer reviewer, grant reviewer — produced qualitatively distinct feedback: instructors were encouraging and process-oriented, peers supportive and conversational, grant reviewers formal and outcomes-oriented. These differences were epistemic, not merely stylistic, foregrounding different aspects of design practice — evidence that role-aware prompting can generate role-sensitive evaluative feedback rather than a single generic response.
-
The division of labor is stage-specific. In a randomized design-thinking experiment, GenAI feedback produced larger gains at the empathize stage, where students valued timeliness and privacy, while instructors produced larger gains at prototype, where contextual anchoring mattered (Fei, 2026).
-
Teacher feedback can be classified empirically into ten types. Lee et al. (2025) coded 100 teacher-authored feedback messages on open-ended middle school math responses with six expert coders. A 12-factor Minimum Rank Factor Analysis with Promax rotation over 108 non-null variables and silhouette-selected K-means clustering yielded 10 types - Direct Correction, Targeted Affirmation, Hinting, Non-verbal Affirmation, Rhetorical Question, Show Your Work, Generic Comment, Direct Critique, Worked Example, and Non-specific Positivity. Prevalence ran 26, 16, 4, 2, and 1 observations across the largest clusters.
-
Feedback literacy and calibration are separable levers. With 120 undergraduates, a Feedback Literacy Script (Filter–Reason–Act–Check) raised writing quality, effective uptake, and deep revision, an Assessment-Performance Calibration Activity improved self-assessment accuracy and cut overconfidence, and combining them produced the largest writing gains and the strongest retention after AI support was withdrawn (Dai (2026)).
-
Clarity is not effectiveness. LLM-rewritten Python error messages were rated significantly more readable and less cognitively demanding than standard interpreter output, yet produced no significant gain in fix rate, attempts, or time-to-fix (Moraru, Biswas & Gadiraju (2026)).
What makes feedback effective: calibration evidence
Ngai & Gilbert (2026) provide a clean experimental result on feedback design with direct relevance to AIED: veridical, immediate, trial-by-trial feedback tied to a learner's own prior prediction is what changes behavior — prediction or beliefs alone are not enough. In their four-group design, feedback combined with a preceding performance prediction improved calibration and behavior, while predictions without feedback did nothing, and adding an explicit over-/underconfidence label added nothing further. LLMs can also serve as calibration partners in this process: Yaşar et al. (2026) found that after rubric refinement, LLMs demonstrated greater consistency than some human raters in applying performance thresholds — useful for norming sessions and formative peer feedback environments, where the model helps calibrate evaluative judgment rather than replace it. This aligns with the feedback-system view: feedback's power lies in closing the gap against a learner's own estimate (the provision–uptake pairing), and effective feedback should be immediate, accurate, and explicitly connected to what the learner predicted — a design principle for AI feedback systems and feedback literacy training alike.
Effort, however, is not automatically the missing ingredient. In a preregistered between-subjects experiment with 302 US adults learning introductory Python, Asher, Gold & Carvalho (2025) paired personalized Claude Sonnet 3.5 explanatory feedback with a structured three-step self-explanation prompt — for each incorrect code section, what the correct code does and why the learner's own attempt failed — and the added reflection bought nothing. Reflective Practice participants spent 4.1 minutes reviewing feedback against 2.1 minutes in plain Practice (t(267) = 11.30, p < .001), which left room for only 2.0 practice problems instead of 3.4 (t(267) = 8.20, p < .001). Each problem was worth the same either way (Reflection × problem-number OR = 1.03, z = 0.23, p = .486), so Practice ended with higher mastery, 79% versus 65% (d = .41), and never fell behind on near-transfer, far-transfer, or code-evaluation items. Reflection also failed to deliver the metacognitive benefit it targeted: judgments of learning did not differ across arms, and the practice conditions were better calibrated (underconfident by 11 points, against video watchers 20 points overconfident, d = −.93) and reported less distraction (d = −.80) than reflective learners, whose reflections averaged only 35 words. The authors' reading is that the AI feedback had already supplied the scaffolding that self-explanation normally adds — a message that names the error, explains the correct approach and describes the underlying concept leaves little for a reflection prompt to do. Writing reflections while still learning unfamiliar syntax also competed directly with the varied repetition that builds transferable skill. This locates a boundary condition for desirable difficulties and productive failure on the processing-depth axis: effort pays when it drives additional retrieval, application or comparison, and is redundant when the learner is already receiving a personalized explanation of their specific error. It also bears on how feedback quality interacts with uptake. The more elaborated and targeted the feedback, the narrower the space for an added layer to add value, which shifts design attention from deepening one episode to spending scarce practice minutes on volume, problem variability, and timing reflection to moments such as a detected plateau. The study tested a single unvetted enhancement, and the authors note that these variants remain open.
- Focus and adaptivity are separable dimensions of feedback. Annotation of teacher and LLM comments in three university writing courses with seven feedback focus types shows that most models cover nearly every type while none matches the teacher's distribution across them, and none shifts its focus across draft stages the way an expert instructor does (Almousa et al., 2026).
- Aggregate gains can hide offsetting shifts. Three months of WISE Agent feedback with 260 Chinese sixth-graders moved no total critical-thinking disposition score (t=-0.854, p=0.394), yet Truth-seeking rose (p<0.001) while Cognitive Maturity fell (p=0.019) — a dip the authors read as students recalibrating overestimated understanding. (Zhu et al. (2026))
Feedback across assessment contexts
The feedback research spans the full range of assessment contexts, each with distinct feedback dynamics:
-
Formative feedback is the canonical site of feedback-for-learning — formative assessment feeds self-regulated learning through feedback that arrives while learning is still in progress (automated formative assessments, feedback enactment).
-
Summative feedback is the feedback attached to summative assessment — end-of-unit tests, oral exams, and proctored/closed-book examinations. In the AI era, the summative setting is where AI resistance matters most (see summative assessment): feedback on a proctored oral exam or closed-book assessment tests genuine learning, whereas feedback on AI-assisted homework can be inflated. Oral assessments reframe the feedback moment as a live, interactive dialogue — feedback becomes immediate, conversational, and inseparable from the assessment itself, which is precisely why they resist AI substitution.
-
Peer feedback adds a social layer — peer feedback develops feedback literacy and is increasingly AI-augmented (AI peer feedback, GenAI in EFL peer feedback). Yet peer feedback is only as good as its actual delivery: in a semester-long field experiment, under two-thirds of students assigned peer feedback received no textual feedback at all, and much of what was delivered was non-targeted praise — so when individual GPT-4 feedback replaced unreliable peers, students sustained the highest participation and posted the largest content learning gains (Geschwind et al., 2026), evidence that the AI's edge is a reliability advantage rather than inherent superiority.
-
Prompt scaffolding turns GenAI into a feedback facilitator. Chang et al. (2026) found that students given structured prompts (context, objective, role) for peer feedback produced more explanations and suggestions, and only that group excelled on rebuttal data and on addressing the opposing view.
-
Automated feedback scales delivery — automated scoring and feedback from essay scoring to short-answer grading. Arthur (Yin et al. 2026) extends this to unstructured quantitative problems: a per-question XGBoost diagnosis backbone infers rubric-labeled mistakes from students' submitted numerical answers to Engineering Economics Calculated Formula Questions and turns them into natural-language feedback, while a dialogue-based scheme requests intermediate answers only when confidence is low — balancing feedback accuracy against the collection burden on students.
-
Automated writing evaluation is a delivery-side tool with a measurable cost. A review of 50 empirical studies (2019–2024) found that automated writing evaluation improved grammar, punctuation and sentence construction in 25 studies and reduced teachers' mechanical correction workload in 15, freeing time for argument, coherence and organization. Gains held only where the tools sat inside structured drafting and revision cycles rather than as stand-alone aids. Heavy reliance on them was linked to weaker critical thinking and creativity, and the tools reproduced standardized English norms that EFL users reported as misaligned with their rhetorical traditions (Gres et al., 2026).
A key cross-context insight is that the reliability of feedback depends on the integrity of the assessment it is attached to: feedback is only as trustworthy as the measure it responds to. In the AI era this pushes educators toward authentic and AI-resistant summative formats where the feedback a student receives reflects genuine learning rather than AI-assisted output.
The "assessment for learning" paradigm reframes feedback as an overarching philosophy rather than a component. Mesny, Roberge-Maltais & Galy (2026) synthesize the wider higher-education literature to frame formative, ongoing, and individualized feedback as central to that philosophy, balancing formative with summative purposes and foregrounding student agency, self-regulation, and metacognitive skill. They note, however, that in management education feedback-rich practices remain unevenly adopted: self- and peer-assessment (a site of peer feedback) dominate the literature, while reassessment — which uses feedback-driven second chances to improve learning — is virtually absent.
- Demonstrated actionability is not enacted actionability. In a scoping review of 421 studies applying NLP to open-ended student evaluation of teaching comments, 61.3% produced output at the level of a demonstrated result or stronger, but only 11.6% showed evidence that intended users evaluated or used it (Eicher & da Silva, 2026).
The provision-uptake pairing
The core feedback insight is that feedback quality and feedback literacy are two sides of one system: high-quality feedback is inert without a literate recipient, and a literate student gains little from poor feedback. AI feedback quality covers the provision side (is the feedback accurate, timely, actionable?), while feedback literacy covers the uptake side (can the student judge and act on it?). The feedback loop is what connects them — the mechanism by which quality feedback, received by a literate learner, closes the gap. Designing effective AI feedback therefore means designing both the system and the student.
Corbin, Tai and Flenady (2026) place mutual recognition beneath that pairing: each party must grant the other standing before quality or literacy can operate, so the same comment guides one student and vanishes for another.
Which student-side capacities matter is empirically separable: among 1,507 secondary students who each received one standardized automated feedback message on an argumentative text, only behavioral engagement with feedback and instrumental attitudes towards it predicted revision performance — students at the 75th percentile on either dimension improved by about 0.10 or 0.11 WLE points more than comparable students at the 25th, roughly the 46th to the 54th percentile — while global receptivity, cognitive engagement and experiential attitudes were unrelated to it and perceived usefulness mediated no dimension's path to performance. Usefulness worked on the other outcome instead: global receptivity's effect on situational interest ran entirely through perceived usefulness (indirect β = 0.09, 95% CI [0.06, 0.13]), so revision gains tracked durable habits of acting on feedback while interest tracked the appraisal of a particular message. (Jansen et al., 2026)
A source can differ in how it calibrates, not how well it scores: in a randomized experiment, chain-of-thought AI feedback outscored an experienced teacher on rubric quality yet produced no greater revision gains, while the teacher's feedback stayed calibrated across drafts and GenAI feedback tracked the strength of the initial essay (Farrokhnia et al. (2026)).
How much a student believes about the source changes what they do with identical words. In a three-condition experiment, 148 adults received feedback written by one model about their own p5.js code, labeled either AI-generated or reviewed by a teaching assistant, with delivery held at 60 seconds. Those who believed a human had read their work stayed on task longer (948 s against 741 s, d = 0.61) and kept the tutorial window in focus longer (850 s against 662 s, d = 0.56), while helpfulness ratings did not move and code quality did not improve. Nearly half of the human-attributed group disbelieved the label, and those participants wrote shorter, less complex programs than the openly AI-labeled group. Source credibility is part of uptake, and it is fragile: a misattributed source can depress engagement below plain disclosure (Morris & Maes, 2026).
A two-layer model shows how the pairing can be organized around a machine without delegating judgment to it. In El Khoury and Ma's worked example, an agent produces a draft rubric-based evidence report on each transcript with every rating tied to quoted excerpts, and the instructor then reviews the report, leads a debrief and decides what the evidence means for that student — a division the authors summarize as the AI organizing evidence while the instructor interprets it. Their claim about uptake is appraisal-based: feedback and iteration outside the social hierarchies students navigate with peers and instructors are easier to attempt, and rehearsal at the student's own pace is what they argue turns occasional confidence into a settled habit of engaging with feedback.
Uptake capacity conditions what AI feedback yields. Across 486 EFL students, the link from AI feedback use to autonomy was 0.48 at one SD above the modeled AI literacy mean and 0.23 below it, with a 0.10 serial indirect path on to deep cognitive engagement (He et al. (2026)).
Why feedback matters for AI in education
Feedback is one of the most consequential and best-evidenced mechanisms in education, and AI both amplifies and complicates it. Well-architected AI feedback can match or exceed human feedback and scale across cohorts, but it demands new learner capabilities (feedback literacy, AI literacy) and carries risks (uncritical acceptance, Over-Reliance). As AI-generated feedback becomes ubiquitous, the lesson is to treat feedback as a whole system — quality, loop, literacy, and assessment context working together — rather than as any single component.
-
Feedback can target the feedback-giver. Tutor training is an under-explored feedback context: automated systems that sequence-label tutors' open-ended responses to flag praise types (effort-based praise — the desirable process-focused kind — vs. outcome-based vs. person-based, the least desirable trait-attributing kind) can deliver explanatory feedback that changes tutoring behavior. The binding constraint is expert-labeled data. An augmentation approach — GPT-4o generates synthetic labeled responses expanding a 13-instance low-resource set to 520, then fine-tunes GPT-3.5 — recovers most of the gap: F2 +20.5% on effort-based praise, +17.3% on outcome-based, and +16.5% on the unseen person-based category. The generalization test to an unseen praise type is the validity check that separates this from overfitting, and the mechanism is closer to teacher-model distillation than classical augmentation. (Xu et al., 2025)
-
Feedback that cannot see the revision gets repeated. In a science learning game whose agents critiqued draft arguments, students revised across multiple rounds, but a quantized 8B model returned substantially the same critique to an improved draft, so the loop that motivated revision also produced the frustration that ended it (Humburg et al., 2025).
-
"Tone is a detail of feedback quality." In the benchmark it tracked the number it was given: every model, told a grade that contradicted the work, wrote feedback whose tone followed that grade, and for eight of 22 the shift per point matched or exceeded the shift with the work's quality. Only the tone rule was validated against trained annotators, on 1,655 tutor replies, so substance is unmeasured. A grader that writes feedback from a wrong grade launders the error into praise or criticism.
Feedback can be built from the process rather than the product. In Xian et al. (2026) one arm received personalized visualizations of its own keystroke and screen activity with structured reflection, while the other received written instructor comments on accuracy, clarity and editorial quality. The process arm gained 5.64 points on a blind 0-100 editorial rating against 1.91 for the product arm, with a Group by Time interaction (F(1, 20) = 4.94, p = 0.038, partial eta squared 0.198) and the control arm's own gain not significant (p = 0.123). Eleven students per arm means the direction is the finding, not the size.
-
Demand for AI feedback can measure scarcity, not preference. In an Honors course on grading, students welcomed AI-generated feedback mainly because they had received so little instructor feedback, not because they judged it better, and they wanted it bounded by instructor judgment (Guberman, 2026).
-
Human-vetted AI feedback improved revisions but not post-test learning. In a randomized trial with 354 undergraduates and 1,366 essays, AI-mediated TA feedback raised revision quality (Cohen's d = 0.50) while post-test scores stayed flat, and the gain grew as TAs adopted more suggestions (Lu et al., 2026).
Two new studies narrow what AI feedback can claim. Across 391 German secondary students, no outcome differed significantly between teacher-written and LLM feedback: revisions, later writing, time on task and perceptions were statistically indistinguishable. Bayesian posteriors leaned toward teacher feedback on later writing (.80) and affect (.90), with a quarter to two-fifths of the mass inside the equivalence region. Immediate AI feedback alone was rated more useful and more motivating (Meyer et al., 2026). In an eight-week study of 247 undergraduates, perceived feedback quality predicted competence (β = 0.52), autonomy (β = 0.44) and relatedness (β = 0.28), which locates part of the effect in how feedback is experienced rather than in its accuracy alone (Liu et al., 2026). The difference is dimensional rather than overall. In a crossover experiment with 36 pre-service teachers, teacher feedback produced larger gains in writing teaching objectives while AI feedback produced larger gains in designing the teaching process, and teacher feedback led on every flow dimension (Wang et al., 2026).
The L2 synthesis gives that hierarchy an empirical anchor. Across 48 controlled studies, GenAI deployed as a feedback and assessment provider produced the largest effect (g = 1.15, falling to g = 0.83 after outlier removal), ahead of its use as an interaction partner (g = 0.81) or an instructional integrator (g = 0.28) (Zhang et al., 2026).
Common Misconceptions
- "Students cannot tell AI feedback from human feedback." They can, and the label changes the judgment: once the label was revealed, ratings of the AI variant fell and of the human variant rose — significantly for Genuineness (p < .01). Provider credibility shaped the evaluation of identical-style feedback.
- "Lab approval predicts semester use." Lab approval is not semester use: all 18 students reacted positively to the GenAI explanation function in lab sessions, yet only half used it during the term. The intention-usage gap is the design problem.
- "Immediate corrective feedback is solved with LLMs." Generating it inside a tutor is feasible but imperfect: across 6,926 logged transactions in the Apprentice Tutor collection, the feedback was good enough to deploy and not good enough to assume correctness.
Connected Concepts
Connected Articles
- Same Feedback, Different Source: How AI vs. Human Feedback Attribution and Credibility Shape Learner Behavior in Computing Education
- Comparing Teacher and AI-Generated Feedback in the Writing Classroom: Experimental Results from Secondary School Classrooms
- AI-Assisted Feedback in Higher Education: Mechanisms, Efficacy, and Motivational Pathways — A Mixed-Methods Investigation
- AI-Mediated Feedback Improves Student Revisions: A Randomized Trial with FeedbackWriter in a Large Undergraduate Course
- Control or Support: Early Student Perceptions of Generative AI in a Course on Grading
- SciStory: Designing AI-Supported Inquiry in Science Learning Games
- Automated feedback on argumentative writing: The role of secondary students' feedback receptivity and feedback perception
- AI Agents, Joyful Assessment, and Third Space: Rethinking Assessment in the GenAI Era
- Who Should Grade My Work? Student Perspectives on Transparent AI-Assisted Writing Assessment in Higher Education
- Who grades best? Comparing ChatGPT, peer, and instructor evaluations across varying levels of student project quality
- Layer-sensitive cognitive offloading in generative AI-assisted writing: Supported performance and independent no-AI outcomes
- Fostering self-regulated learning through adaptive learning technology: A differentiated perspective on the role of feedback
- Generative AI offers more, but students revise less: comparing the effects of teacher and AI feedback on student essay revisions
- Supporting self-regulated learning through generative AI feedback in online higher education: the importance of student perceptions of the source of feedback
- Leveraging generative AI to facilitate peer feedback in collaborative argumentation learning
- Is It Ethical for Teachers to Use AI for Student Feedback?
- Metacognitive training facilitates optimal cognitive offloading
- AI feedback literacy in higher education: Understanding, measuring, and predicting student feedback uptake
- Generative artificial intelligence as an enabler of student feedback engagement: a framework
- Making sense of AI feedback: how students' feedback literacy moderates the link between ChatGPT acceptance and self-regulated learning
- Rethinking AI-assisted writing instruction: feedback literacy scripts, calibration training, and student writing development
- Students' Perceptions of Generative AI-Powered Learning Analytics in the Feedback Process: A Feedback Literacy Perspective
- Reconnecting relationships through technology: Developing feedback literacy capabilities through an AI automated feedback tool
- Positioning generative AI in EFL peer feedback: Training feedback literacy and enabling uptake in speaking classes
- Artificial intelligence and feedback in university education: effectiveness and student perceptions
- The care-full craft of feedback in an age of generative AI
- How artificial intelligence transforms the feedback ecosystem in higher education
- Enhancing learner-centered feedback with AI: Teachers' practices and perceptions
- Human-centered GenAI feedback design in higher education: A multisite experiment on direct, reflective, and hybrid approaches to scientific argumentation
- Using AI-Generated Feedback to Improve Critical Thinking and Writing Proficiency
- Using generative AI to promote psychological, feedback, and artificial intelligence literacies in undergraduate psychology
- LearnLens: LLM-Enabled Personalised, Curriculum-Grounded Feedback with Educators in the Loop
- Assessing the Impact and Underlying Pathways of Sequenced AI Feedback on Student Learning
- AICoFe: Implementation and Deployment of an AI-Based Collaborative Feedback System for Higher Education
- The Effect of High-Frequency, Automatically-marked Formative Assessments on Student Outcomes in A-Level Sciences
- Reconsidering the use of oral exams and assessments: An old way to move into a new future
- Marked pedagogies: Examining linguistic biases in personalized automated writing feedback
- From scoring to explanations: Evaluating SHAP and LLM rationales for rubric-based teaching quality assessment
- Beyond the Traceback: Using LLMs for Adaptive Explanations of Programming Errors
- Students' Perceptions of Artificial Intelligence Tools for Study Productivity and Learning: An Exploratory Survey Study
- From evaluation to emulation: LLMs as agents of iterative pedagogical design
- GPT-4 feedback increases student activation and learning outcomes in higher education
- Generating In-Context, Personalized Feedback for Intelligent Tutors with Large Language Models
- Arthur: An artificial intelligence powered teaching assistant system for Engineering Economics class
- Innovative assessment and grading practices in higher education: A critical exploration for management educators
- Can ChatGPT Replace the Teacher in Assessment? A Review of Research on the Use of Large Language Models in Grading and Providing Feedback
- A Tripartite Feedback Framework for AI-Assisted Assessment of Complex Reports in Higher Education
- Benefit or Bottleneck? Assessing the impact of structured reflection on learning from AI-driven explanatory feedback
- Bounded reliance: A source credibility perspective on EFL students' engagement with AI-generated writing feedback
- From Content Generation to Learning Support: Pedagogy-Guided Generative Video Tutors for STEM Learning
- Evaluating feedback focus and pedagogical adaptivity in LLM-generated feedback on student writing
- From Sentiment Classification to Actionable and Responsible Feedback: A Scoping Review and Evidence Map of NLP in Student Evaluation of Teaching, 2015–2026
- Who Gives Feedback Matters: Student Biases Towards Human and AI-Generated Formative Feedback
- AI-Powered Feedback System for Writing: A Review of Automated Writing Evaluation (AWE) Tools
- Human-written and AI-generated revisions: An appraisal analysis of dialogic positioning in argumentative writing
- Improving automated feedback systems for tutor training in low-resource scenarios through data augmentation
- A Conversational AI Framework for Teacher Training
- Trade-offs in assessment and feedback design: a trilemma-based model
- Beyond agreement: a benchmark of how language models grade student work, and how they fail
- As effective as the teacher, but less motivating: A randomized controlled comparison of AI-generated and teacher feedback on students' writing performance and intrinsic motivation
- Process visualization feedback in news translation: a controlled intervention study of human-AI interaction using multimodal process data
- When Does GenAI Work for L2 Learning? A Three-Level Meta-Analysis
- Comparing human-AI collaboration and teacher-student collaboration in developing pre-service teachers' instructional design competence: an experimental study
Connected FAQs
Connected Resources
- ClarityAn open-source Agent Skill and private browser editor that turn eighteen rules for clearer writing into draft, rewrite and review modes for prose.
- The Pedagogical PromptbookAn open-access book of evaluated prompts for evidence-based teaching and instructional design practices, each chapter publishing the prompt, its development process and its evaluation results.
- SnorklA classroom tool that gives students instant AI feedback on recorded spoken and drawn explanations of their thinking, and gives teachers analytics on those explanations.