Feedback¶
Element · All elements
Evidence · 187 claims (83 for, 89 mixed, 15 against) · 142 studies (54 causal, 27 review, 20 quant-synthesis, 15 associational, 9 qualitative, 9 design, 8 theoretical),q1–q4· 36 of 142 report an effect size · 164 claims rest on one study
Description¶
Feedback is information provided to learners about their performance or understanding, intended to close the gap between current and desired performance. Effective feedback answers three questions: Where am I going? How am I doing? Where to next? (Hattie & Timperley, 2007). It functions as the corrective mechanism in any instructional cycle that includes Practice or Assessment.
Design Implications¶
Feedback is among the most powerful influences on learning, but its effects are highly variable — it can improve or degrade performance depending on design Feedback effects on performance are highly variable and can be negative. [~S]. Feedback should focus on the task and the process, not the person; praise directed at the self carries little informational value and can undermine motivation. Feedback is most effective when learners have an opportunity to act on it — feedback delivered after grading, or without a revision opportunity, is largely wasted Formative feedback with opportunities to act improves achievement. [+S].
Context¶
Requirements¶
- Clear success criteria so learners can interpret the feedback against a target
- Timing that lets learners apply the information while the task is still active
- Specific, actionable comments tied to the task or strategy, not global judgments
- A revision or retry cycle — feedback without a subsequent action is inert (Formative Assessment)
Constraints¶
- Feedback on performance level alone (grades, scores) without process information adds little and can reduce performance for struggling learners Feedback effects on performance are highly variable and can be negative. [-S]
- Overly detailed feedback can overwhelm novices; too much guidance at once increases extraneous load [~M]
- Feedback that threatens the ego (public correction, comparative ranking) triggers disengagement rather than improvement [-M]
- Immediate feedback is not always superior — delayed feedback can benefit retention for some tasks [~W]
Target Learners¶
- All learners, but the kind of feedback should shift with expertise: novices benefit from task-level corrective feedback, while advanced learners benefit more from process-level prompts that trigger Self-Explanation Self-explanation improves conceptual understanding. [+M]
- Struggling learners need feedback framed around process goals rather than outcome goals Process goals outperform outcome goals for novices. [+M]
- Learners monitoring their own progress benefit when feedback builds self-assessment capacity Self-monitoring improves self-regulation. [+M]
Target Learning Goals¶
- Knowledge acquisition: correcting misconceptions and errors close in time to performance
- Skill refinement: iterative improvement of procedural and compositional work
- Self-regulation: developing learners' capacity to evaluate and adjust their own work
Affordances¶
- Feedback Loops — feedback is the mechanism that closes the loop between performance and adjustment; the design question is loop latency (how quickly learners learn whether their attempt worked)
- Mastery Learning — mastery models are unworkable without feedback; learners must know what remains unmastered and receive targeted input before re-attempting
- Assessment for Learning — feedback converts assessment from measurement into instruction; the assessment itself becomes the feedback event
- Scaffolding — feedback is a responsive scaffold that fades as learners internalize evaluative criteria Fading support promotes transfer of responsibility. [+M]
Claims¶
- Feedback improves learning by a medium amount on average, but its effect varies widely with the information it carries, and a substantial share of feedback effects are negative [+S] — (also cited [~M])
- Feedback Most Effective At Task And Process Levels [+S]
- Praise for intelligence, rather than for effort, after success undermines children's motivation and performance after a later failure [-M]
- Feedback Enhances Retrieval Practice [+S]
- Retrieval practice improves long-term retention [+S]
- Taking initial multiple-choice tests without feedback can lead students to later produce the incorrect lure answers they selected, even when an overall retrieval practice benefit occurs [-M]
- Providing feedback after initial multiple-choice tests cut lure intrusions on final short-answer tests roughly in half, with immediate and delayed feedback equally effective [+M]
- Initial short-answer tests outperform initial multiple-choice tests mainly when feedback follows them; without feedback, the higher initial success of multiple-choice tests can favor multiple-choice [~M]
- High-confidence errors are more likely than low-confidence errors to be corrected on a later test after feedback. [+W]
- Incidental vocabulary gains show retention problems, with losses over time moderated by feedback timing and topic interest [~W]
- Peer Feedback Improves Writing [+M]
- Peer assessment improves academic performance by a small average amount relative to no assessment or to teacher assessment [+M]
- In two secondary-school writing studies, peer reviewers' own performance predicted how helpful their feedback was but not how accurately it identified problems [~M]
- In two online comparisons, a rubric changed what peer reviewers commented on (one study) and helped peer raters place writing in quality tiers, most for complex criteria (the other); the first study says the rubric restrained reflective thinking [+W]
- Students find computer-based simulation feedback insufficient because incorrect responses are marked without explanation [-W]
- Assessment for learning improves achievement [+M]
- Self Assessment Accuracy Is Low Without Training [~M]
- Corrective feedback on second-language errors has a significant effect on L2 development that is maintained on delayed tests [+S]
- The literature is supportive of recasts, but recast effectiveness varies with the language area and learner knowledge, focus, and aptitudes [~M]
- Metalinguistic feedback outperforms recasts on the immediate post-test, but the two do not differ significantly on the delayed post-test [~M]
- Metalinguistic feedback's gains on wh-question performance decline over a ten-day delay while recast gains hold, leaving the two no different [~W]
- Learners who could decide after a trial whether to receive knowledge of results (Self-After, Self-Both) were more accurate on a 24-h motor retention test than learners who decided before the trial (Self-Before) and than their yoked counterparts. [+W]
- Having control over the knowledge-of-results decision only before a motor trial (Self-Before) produced no statistically significant retention or transfer benefit over a yoked group receiving the same KR schedule. [~W]
- Simulation Based Education With Deliberate Practice Improves Clinical Outcomes [+M]
- Simulation Based Education Improves Outcomes [~M]
- Human tutoring has a medium effect over no tutoring, and in one small study students learned as well when tutors only prompted them as when tutors also explained and gave feedback [~W]
- Generative AI-assisted feedback is associated with a statistically significant pretest–posttest improvement in violin students' intonation performance over a four-week period [+W] — attached 2026-10-10 from Yakup Aksoy (2026), which proposed "AI-assisted intonation feedback model: Cubase cent measurement analyzed by ChatGPT and DeepSeek without teacher intervention"; tests this page's relationship.
- Students reported satisfaction with the detailed analyses, personalized multi-dimensional recommendations and motivational guidance provided by the AI systems [+W] — attached 2026-10-10 from Yakup Aksoy (2026), which proposed "AI-assisted intonation feedback model: Cubase cent measurement analyzed by ChatGPT and DeepSeek without teacher intervention".
- AWE tools such as Grammarly, Quillbot and DeepL Write significantly improve writing precision, fluency and grammatical proficiency through immediate and customized feedback [+W] — attached 2026-10-10 from Gres et al. (2026), which proposed "Automated Writing Evaluation (AWE) tools: Grammarly, Quillbot and DeepL Write"; tests this page's relationship.
- AI multimodal feedback produces post-test learning gains equivalent to educator feedback after controlling for pre-test performance [~M] — attached 2026-10-11 from Chloe Qianhui Zhao et al. (2026), which proposed "Real-time AI-facilitated multimodal feedback system integrating structured text, retrieved slide pages, and streaming audio narration"; tests this page's relationship.
- AI multimodal feedback is rated higher than educator feedback on specificity, clarity, simplicity, motivation, satisfaction, and lower mental effort, with comparable correctness, trust, and acceptance [+M] — attached 2026-10-11 from Chloe Qianhui Zhao et al. (2026), which proposed "Real-time AI-facilitated multimodal feedback system integrating structured text, retrieved slide pages, and streaming audio narration".
- Higher education students' interactions with an AI automated feedback tool can support the development of their feedback literacy capabilities [+W] — attached 2026-10-12 from Laura Tubino (2022), which proposed "AI automated feedback tool piloted university-wide for student feedback".
- Customized AI agents generate feedback with higher accuracy and specificity than general-purpose agents in self-directed instructional design tasks [+W] — attached 2026-10-12 from Han Xue et al. (2026), which proposed "Customized vs. general-purpose LLM-based educational agents on the Coze platform"; tests this page's relationship.
- The customized agent's advantage in short-term task improvement emerges under directive feedback but not under Socratic feedback [~W] — attached 2026-10-12 from Han Xue et al. (2026), which proposed "Customized vs. general-purpose LLM-based educational agents on the Coze platform"; tests this page's relationship.
- Plan quality improves more after LLM-generated adaptive feedback than after high-quality generic guidance [+W] — attached 2026-10-12 from Sukjaitham et al. (2026), which proposed "Quality-matched feedback design: LLM adaptive feedback versus expert generic guidance materials".
- Equivalence analyses indicated practical comparability between GPT-o4-mini and teacher feedback, while DeepSeek R1 demonstrated non-inferiority to teacher feedback [~W] — attached 2026-10-12 from Grion et al. (2026), which proposed "AI-generated feedback from GPT-o4-mini and DeepSeek R1 as formative assessment components in a project-based university course"; tests this page's relationship.
- Students' perceptions of mastery, emotions, and satisfaction were similarly high across all three feedback conditions [~W] — attached 2026-10-12 from Grion et al. (2026), which proposed "AI-generated feedback from GPT-o4-mini and DeepSeek R1 as formative assessment components in a project-based university course".
- In the expectations meta-analysis, the amount of feedback agents provided had only a meager relation to performance (r = .07) whereas the climate agents created had a strong one (r = .36) [-W]
- AI-powered corrective feedback and repeated practice enhance oral language skills (Communicative Capacity Pathway) [+W]
- AI-peer integrated feedback produces significantly higher posttest behavioral engagement than conventional peer feedback in EFL writing [+W]
- In one quasi-experiment in EFL writing classes, students given AI-peer integrated feedback reported higher posttest cognitive engagement than students given conventional peer feedback [+W]
- Students given AI assistance in peer feedback tended to rely on the AI-generated feedback and struggled to perform the task when the assistance was removed [~W]
- Undergraduate students' preferences for AI feedback characteristics differ significantly from an equal distribution for tone, length, format, and focus, but not for timing [~W]
- AI-supported co-produced feedback achieved comparable quality and statistically significant learning gains with reduced production time and workload [+W]
- Business Writing students valued AI feedback for higher-order and sentence-level comments but found it sometimes too general, while appreciating peers' contextual knowledge [~W]
- AWE feedback has limited depth and questionable accuracy for evaluating complex writing aspects [-W]
- Teachers valued CulturAIEd's demographic integration, customization, and rubric feedback as highly efficient [+W]
- Children perceive LLM-generated adaptive feedback as more helpful and more motivating than generic guidance [+M]
- Detailed written answer keys at checkpoints significantly improved momentum post-test scores over brief answers, but not kinetic energy scores [~W]
- Expert evaluation found the virtual patient realistic and immediate ACT feedback increased therapists' awareness of intervention choices [+W]
- Across tasks, teacher feedback shifted from basic error correction toward higher-order dimensions while AI-assisted peer feedback reduced content suggestions and increased technical-detail feedback [~W]
- Explanatory feedback alone raises both perceived fairness and decision acceptance, and the indirect effect of agent on acceptance is significant only when explanations are provided [~W]
- Direct GenAI-supported feedback improves immediate argument-quality gain over peer feedback, while reflective and hybrid designs outperform direct GenAI feedback [~W]
- GPT-generated feedback passes simulated-student helpfulness evaluations only about 35% of the time on real student data [~W]
- Judge-condition verification did not consistently improve LLM feedback quality and often propagated Peer agents' errors [-W]
- GPT-4 generated detailed per-response feedback that was useful for identifying scoring errors and could support personalized student feedback [~W]
- High-acceptance groups made more effective use of AI feedback during revision, resulting in higher accuracy scores [+W]
- Children's plan quality improves substantially after receiving LLM-generated adaptive feedback [+W]
- Revision after chatbot feedback predominantly results in successful uptake, indicating effective error correction [~W]
- Instructors use more first-person and second-person pronouns and more questions than LLMs in feedback comments [~W]
- No statistically significant interaction effects between satisfaction and adaptive feedback on the subsequent pre-actional phase; only marginal trends emerged (H4 not supported) [~W]
- Trial-level perceived usefulness of feedback is associated with the magnitude of plan quality improvement [+W]
- Proactive multimodal feedback significantly reduces debugging time on task compared to control [+W]
- The most common answer transition after KITE feedback was Partially Correct to Correct, occurring in 14 of 44 cases (31.82%) [~W]
- Low-level feedback automation is appropriate only for sentence-level clarity; argumentative unclarity should trigger human review [~W]
- Live feedback shifts the router's strategy from the Feynman prompt favored in simulation to the scaffolding Coach prompt, with superior live performance [+W]
- In the second assessment stage, ChatGPT-4o feedback was significantly higher than peer group feedback in constructiveness, accuracy and concreteness [+W]
- Timely feedback on homework increases learning, and computer platforms providing immediate feedback help teachers adapt [+W]
- AI companions are frictionless and sycophantic, lacking the corrective feedback that real relationships provide [+W]
- Similarity between AI and peer feedback reassured students and increased their confidence in peer response [+W]
- In the first writing task, teacher feedback contained significantly more items than AI-assisted peer feedback, especially on vocabulary errors and technical details [+W]
- Teacher feedback produced larger first-task score gains that dropped 32.8% by the second task, while AI-assisted peer feedback gains remained stable (a learning-curve effect) [~W]
- Submissions receiving valid LLM-generated test-case feedback had higher pass rates, but valid test cases were not associated with fewer subsequent submissions [+W]
- AI-mediating is a newly observed orchestration type in which the teacher concretizes or supplements correct AI feedback for students in curriculum context (12 of 46 episodes) [~W]
- Qualitatively, AI-peer integrated feedback was associated with a broader repertoire of revision operations and more agentive cognitive strategies than peer feedback alone [+M]
- AI-peer integrated feedback produces significantly higher posttest affective engagement than conventional peer feedback in EFL writing [+W]
- AI-peer integrated feedback yields significant gains on all four writing dimensions, with the group × dimension interaction showing differential improvement favoring task achievement and lexical resource over coherence and cohesion [+M]
- AI-assisted feedback drafts increase total feedback length by 39.79 characters per submission [~W]
- Distance students prefer an institution-embedded assistant offering 24/7 immediate, context-specific feedback over generic content such as ChatGPT [+W]
- Automated formative assessment and feedback with AI agents ranks second-highest in impact with moderate perceived challenge [+W]
- Both ChatGPT feedback and peer feedback prompted doctoral students to actively exercise hard evaluative judgement in statistical analysis, through different challenges [+W]
- In a small case study, students given ChatGPT feedback valued its immediacy and alternative analytical perspectives, but one distrusted its answers as 'often made up' for complex data analysis [~W]
- Students trust educator feedback more than AI-generated feedback: 90.5% versus 60.1% rated it somewhat or very trustworthy in an Australian survey [~W]
- Feedback channel is a non-significant moderator of feedback effects [~W]
- Students who continued feedback conversations with ChatGPT perceived more utility in AI feedback than one-shot users [+M]
- Feedback has a medium overall effect on student learning (d = 0.48 after outlier removal) with substantial heterogeneity [+M]
- Students aware of the GenAI feedback source reported no attitude change, while unaware students varied between content-focused acceptance and conditional willingness to reconsider [~W]
- Generative AI-generated hints can be as effective as human-generated hints for mathematics performance, but ensuring accuracy requires validation procedures that discard a substantial share of generated hints [~W]
- ChatGPT and DeepSeek feedback differed from student to student in technical content, repertoire and address style, providing an individualized approach [+W]
- Feedback type shifts engagement by question type: educator feedback yields more MCQ submissions while AI feedback trends toward more open-ended revisions, with no overall difference [~W]
- GenAI can provide personalised feedback in group work, but the depth of personalisation is contested [~W]
- Students rate human-generated feedback as superior to AI-generated feedback in quality, helpfulness, and clarity, and as more likely to be acted upon [+W]
- GenAI-supported peer feedback groups achieved higher structural complexity of argumentation artifacts than peer-feedback-only groups [+W]
- GenAI feedback quality correlates with initial essay quality while teacher feedback quality does not [~W]
- Gender prompts shift feedback stance: female students receive affective, first-person, empathy-oriented feedback while male students receive objective, task-oriented critique [~W]
- AI-assisted peer feedback enhanced lexical diversity only in the first writing task, with the group advantage disappearing by the second task [~W]
- Hybrid feedback outperforms direct GenAI feedback on conceptual learning; the reflective–direct contrast is positive but non-significant [~W]
- LLM-generated feedback is more comprehensive in volume than human feedback but is sometimes too generic, abstract, or misaligned with the assigned grade [~W]
- The total effect of layered feedback on learning outcomes was significantly negative, with an unmediated negative direct effect [-W]
- Test-takers rate LLM-generated feedback as persuasive and useful as human instructor feedback (Wilcoxon signed-rank tests show no significant difference) [+W]
- All three LLM-generated feedback types are associated with faster time to solution than a compiler-errors-only baseline, with Single-Issue Feedback showing the largest reduction [+W]
- Non-layered AI feedback produced significantly higher post-test performance than layered AI feedback after controlling for pre-test [~W]
- After PAIRR, most students prefer combined peer and AI feedback over either source alone [+W]
- Most individual argument structure and quality dimensions showed no significant differences across the three peer feedback conditions [~W]
- Teachers whose principals received the program were no more likely to report positive perceptions of the usefulness of the feedback provided [-W]
- Immediate knowledge of results during computerized achievement testing produces essentially no systematic differences in achievement estimates or response dimensionality [~W]
- Proactive multimodal feedback significantly increases feedback uptake compared to control [+W]
- Students' preferences for AI versus human scoring and feedback vary significantly across assessment scenarios, with strong human preference for high-stakes tasks [~W]
- Students perceived peer group feedback as contextually rich but sometimes subjective, and ChatGPT-4o feedback as more objective and well structured but occasionally lacking contextual sensitivity [~W]
- Most students preferred combined peer and AI feedback over either source alone, and rated peer and ChatGPT feedback as similarly useful [+W]
- Students showed no statistically significant apprehension about trusting automated feedback validity, and favoured granular automated marginalia over generalised human feedback [~W]
- Subtle indirect cueing feedback tends to have higher impacts on learning outcomes than other forms of feedback, per research the review cites [~W]
- The effort-based reward reduced how often participants used the assistant, whereas metacognitive feedback changed how they used it [+W]
- Zero-shot GenAI feedback quality is statistically comparable to teacher feedback quality [~W]
- Perceived usefulness of response-level LLM feedback is most strongly associated with actionability and perceived accuracy [+W]
- Descriptive time-to-fix patterns favored pragmatic messages, but statistical tests found no significant effects [~W]
- In the first assessment stage, peer group and ChatGPT-4o feedback showed no significant difference in quality [~W]
- Multi-agent tutoring architectures show superior behaviour modelling and feedback compared to single-agent or rule-based tutors [+W]
- LLM feedback for students marked by race, language, disability, gender, or motivation shows positive feedback bias and feedback withholding bias — more praise and less substantive critique than for unmarked counterparts [-M]
- Students report receiving scores with little individualized feedback and request more in-depth feedback that explains why and how they can improve [+W]
- All three groups benefited from the instructional intervention, with both experimental groups benefiting the most [+W] — attached 2026-10-15 from Ahlem Ammar et al. (2006), which proposed "Four-week form-focused intervention on third-person possessive determiners his and her"; tests this page's relationship.
- Prompts were overall more effective than recasts as corrective feedback for L2 grammar learning [+W] — attached 2026-10-15 from Ahlem Ammar et al. (2006), which proposed "Four-week form-focused intervention on third-person possessive determiners his and her"; tests this page's relationship.
- Trained human evaluators provided higher-quality formative feedback than ChatGPT on four of five components, while ChatGPT outperformed humans on criteria-based feedback [~W] — attached 2026-10-15 from Steiss et al. (2023), which proposed "ChatGPT v.3.5 prompting protocol for generating formative writing feedback".
- ChatGPT feedback quality varied by essay quality: accuracy declined as essay quality rose, prioritization was best for average essays, and supportive tone was lowest for low-quality essays [~W] — attached 2026-10-15 from Steiss et al. (2023), which proposed "ChatGPT v.3.5 prompting protocol for generating formative writing feedback".
- Students generally attempted fewer items under ego-involving than task-involving feedback, but among praised students the gap was not significant [~W]
- Students value GenAI corrective feedback with rationales and report increased motivation and confidence [+W]
- Receiving positive affective feedback from peers is a significant predictor of assessees' project performance [+W]
- Students who received corrective feedback post-processed practice tests more intensely and rated the tests as more useful for monitoring than students without feedback [+W]
- In Experiment 2, neither self-assessment feedback plus correct answers nor plus contrasting own with correct answers improved test-phase self-assessment accuracy [-W]
- Without feedback, more practice-test participation predicts lower specificity (worse detection of incorrect answers); with corrective feedback this negative relation disappears [+W]
- Feedback in full-mission simulator training is delivered through recurring, multidimensional, multimodal feedback cycles built on instructors' embedded assessments of the fit between learning objectives and the ongoing task [+W]
- Comparing self-generated keywords with expert keywords increases bias, specifically underconfidence [~W]
- In Experiment 2, task-selection accuracy during the learning phase did not differ between feedback and control conditions, unlike Experiment 1 [~W]
- In Experiment 1, task-selection accuracy during the no-feedback test phase did not differ between feedback and control conditions [-W]
- A trusting, supportive teacher-student relationship emerged as a prominent motivator for acting on feedback, especially for Information Seekers with Low Emotional Strain [~W]
- Across both experiments, self-assessment feedback failed to improve self-assessment accuracy in the absence of feedback, and learners' self-assessments did not track practice-driven performance gains [-W]
- Within the feedback condition, students rated test taking and receiving feedback as equally useful for monitoring their performance [~W]
- Person praise raises state self-esteem (d = 0.40), and under task-involving feedback higher state self-esteem is negatively related to motivation, supporting a distraction mechanism only in the low-threat condition [~W]
- Pre-service teachers who received NLP-based adaptive feedback were better able to justify their diagnoses than those who received static feedback (reported from Sailer et al.) [+W]
- AI-assisted feedback does not reduce student-perceived usefulness of rated feedback [+W]
- Accurate LLM solution diagnosis does not reliably produce pedagogically actionable feedback, with lowest rubric scores on error identification and actionability [-W]
- AI-correcting is a newly observed orchestration type in which the teacher identifies erroneous AI feedback, points it out to students, and re-guides or edits AI prompts (10 of 46 episodes) [~W]
- Interest in AI for feedback in higher education is driven by workload reduction and feedback timeliness amid massification [~W]
- Students who preferred AI feedback cited clarity, understandability, consistency, specificity, and 24/7 availability [+W]
- Anonymous asynchronous online peer review of writing: lower-skill students sometimes felt inferior yet increased determination, while better writers felt they learned less [~W]
- Acceptance rates increase with writing progress, with the L2-over-L1 advantage widening in later stages [~W]
- The study compared feedback-uptake relationships across two immersion settings totaling 18.3 hr of French and 14.8 hr of Japanese classroom interaction [~W]
- ENL students were near-evenly split in preference between AI-generated and human tutor feedback [~W]
- Feedback source (human tutor vs GPT-4) had no significant effect on ENL students' posttest writing scores [~W]
- GenAI is reshaping assessment practices in computer science education through automation, scalability, and personalized feedback [+W]
- Feedback direction moderates effects: student-to-student feedback is most effective (d = 0.85), teacher-to-student d = 0.47, student-to-teacher d = 0.35 [~W]
- GenAI-supported peer feedback groups produced more suggestions and explanations and less positive affective feedback than the PF group [~W]
- AI-generated (GPT-4) feedback produced no greater linguistic progress than human tutor feedback among ENL students over six weeks [~W]
- Under blind conditions, students rated GenAI feedback more positively than tutor-generated feedback, with a statistically significant difference only on Genuineness favouring GenAI [+M]
- Learners prefer receiving corrective feedback more than teachers feel they should provide it [+W]
- Learners selectively filtered and adopted AI-generated feedback while maintaining pedagogical agency [~W]
- NLP-based feedback generation for textual student answers in large courses reduced grading effort by up to 85% with high precision (reported from Bernius et al.) [+W]
- No significant relationship was found between positive feedback and positive emotions in CPS [~W]
- Praise for intelligence (person praise) induces a fixed mindset and helpless reactions compared with process praise [+W]
- Peer assessment of writing yields outcomes at least as good as teacher assessment and sometimes better [~W]
- Prompt-writing ability is a new feedback-eliciting capacity that shapes GenAI feedback engagement [~W]
- Differences between human and ChatGPT feedback quality were small, with means within one point and partial eta-squared ranging from .01 to .12 [~W]
- Trust in customized agents was associated with clear, accurate, and contextually aligned feedback [~W]
- Teacher roles vary by orchestration type, with content-related and ethical roles prominent when the teacher intervenes in AI-generated feedback, and a knowledge-transmitter role emerging in explain-the-screen [~W]
- Practice with feedback—either correct-answer or explanatory—produces better memory than lecture, in both multiple-choice and open-ended formats [+W] — attached 2026-10-19 from Michael W. Asher et al. (2026), which proposed "GPT-4o-generated personalized explanatory feedback for open-ended practice"; tests this page's relationship.
- Generalization from practice requires explanatory feedback and sufficient prior knowledge or confidence, with benefits concentrated among better-prepared learners [~W] — attached 2026-10-19 from Michael W. Asher et al. (2026), which proposed "GPT-4o-generated personalized explanatory feedback for open-ended practice"; tests this page's relationship.
- Predominantly negative feedback leads students to expect a next round of problems will require more effort than predominantly positive feedback (Experiments 1 and 2) [+W]
- With reduced construct overlap and a no-feedback control, feedback did not significantly affect challenge (Experiments 3 and 4), and negative feedback raised expected effort required only in Experiment 3 [~W]
- Learners judge explanatory feedback as more effective preparation than correct-response feedback, in both studies [+W]
- Correct-answer feedback alone was insufficient for generalization: significantly worse than lecture in Study 1, with the difference not significant in Study 2 [~W]
- Neither content-specific nor generic support during feedback reception shows significant effects on feedback reception quality or subject-matter-related knowledge, but the evidence base is too small to interpret [~W]
- Learning analytics tools can provide personalized feedback on students' academic writing by identifying mistakes and offering constructive suggestions (e.g., AcaWriter) [+W]
- ManyClasses study found no effect of delayed versus immediate feedback or study incentives across 38 classrooms [~W]
- Receiving grades is positively associated with achievement and external motivation compared with receiving no performance feedback (review-reported meta-analysis) [~W]
- Individual feedback impacted perceived utility only to a limited extent, and general feedback lowered perceived utility of retrieval practice (Experiment 2) [~W]
- Students who benefited from retrieval practice rated it as more effective after receiving individual feedback, while other groups showed no statistically significant change (Experiment 1) [~W]
- Feedback targeting reading self-efficacy had a positive effect on motivation, while error-focused feedback and rewards had mixed effects on long-term interest and engagement [+W]
- Learners are more willing to act on feedback when they perceive the sender as credible and trustworthy [+W]
- Curriculum and assessment design factors, including modular structure, feedback timing, and overlap between consecutive assessments, are proposed to moderate feedback use [~W]
- Elaborated feedback did not significantly outperform non-elaborated KR feedback on achievement [~W]
- Adaptive feedback does not improve pre-service teachers' diagnostic accuracy in simulation-based learning [-M] — attached 2026-10-23 from Sailer et al. (2022), which proposed "ANN-based automatic adaptive feedback in simulations of pupils with learning difficulties".
- Under situationally increased self-enhancement motivation, people prefer positive feedback regardless of its perceived accuracy [~W]
Design Decisions¶
What should the feedback say?¶
- Default: tell the learner what was right or wrong and how to improve on the task or the process, not just a score, a reward or praise. Across 435 education studies feedback averaged d = 0.48, rising with information content: reinforcement or punishment d = 0.24, corrective feedback d = 0.46, high-information feedback d = 0.99 — Feedback improves learning by a medium amount on average, but its effect varies widely with the information it carries, and a substantial share of feedback effects are negative [+S], Feedback Most Effective At Task And Process Levels [+S]
- Changes when: the feedback turns attention to the learner rather than the task → expect it to help less or to hurt; more than a third of 607 feedback interventions lowered performance, and effectiveness fell as feedback moved toward the self — Feedback improves learning by a medium amount on average, but its effect varies widely with the information it carries, and a substantial share of feedback effects are negative [-S]
- Changes when: the comment judges the person, discourages, or threatens self-esteem → expect little or harm; in Kluger and DeNisi's moderator data as Hattie and Timperley tabulate it, praise averaged 0.09 against 0.34 without it, feedback designed to discourage −0.14, and feedback under much threat to self-esteem 0.08 against 0.47. Whether feedback is positive or negative mattered less than whether it was about the person or the work, so the main lever is aiming comments at the work and what to do next, more than making their tone warmer — Feedback about self as a person is the least effective feedback level; task, process, and self-regulation feedback support learning and combine for greater impact [-M]
- Changes when: the learner has low self-efficacy → negative feedback is more likely to bring negative affect, less motivation on the next task and attributions to ability, by a review's account of earlier studies with no effect size; one more reason to aim criticism at the work and pair it with a next step — Feedback about self as a person is the least effective feedback level; task, process, and self-regulation feedback support learning and combine for greater impact [~W]
- Changes when: the feedback is praise for ability ("you must be smart") → children praised this way did worse after a later failure than children praised for effort; the harm is moderated by age and gender and is specific to person- or ability-directed praise — Praise for intelligence, rather than for effort, after success undermines children's motivation and performance after a later failure [-M]
- Tested with: school and university learners across subjects (Wisniewski et al., 435 studies); laboratory and workplace tasks as well as education (Kluger & DeNisi); fifth graders for the praise experiments.
- Not settled: the claims cannot separate task, process and self-regulation levels (the 2020 meta-analysis coded information content, not levels, and the four-level ranking is a review's argument resting on between-study moderator contrasts), and no claim here tests whether giving learners a required revision step after feedback changes its effect, although the page's Requirements assume it.
Which responses should get feedback?¶
- Default: make sure errors are corrected; supplying the correct answer after a wrong retrieval raised one-week retention by 494% in a word-pair experiment, while feedback after correct answers made little difference even when learners were unsure — Feedback Enhances Retrieval Practice [+S]
- Default: when practice is a quiz, attach corrective feedback; classroom quizzing with feedback gave g = 0.537 against g = 0.374 without (a between-study comparison; an earlier meta-analysis found no difference), and in the testing-effect meta-analysis g = 0.73 with feedback against 0.39 without — Feedback Enhances Retrieval Practice [+S], Retrieval practice improves long-term retention [+S]
- Changes when: practice uses multiple-choice items → feedback is not optional, because students later reproduce the lures they chose; feedback roughly halved lure intrusions (20% to about 10%), and short-answer practice beat multiple-choice mainly when feedback followed it — Taking initial multiple-choice tests without feedback can lead students to later produce the incorrect lure answers they selected, even when an overall retrieval practice benefit occurs [-M], Providing feedback after initial multiple-choice tests cut lure intrusions on final short-answer tests roughly in half, with immediate and delayed feedback equally effective [+M], Initial short-answer tests outperform initial multiple-choice tests mainly when feedback follows them; without feedback, the higher initial success of multiple-choice tests can favor multiple-choice [~M]
- Changes when: the learner was confident and wrong → correction is more likely to stick (the hypercorrection effect), so confidence-rated items are worth correcting first; no effect size is recorded — High-confidence errors are more likely than low-confidence errors to be corrected on a later test after feedback. [+W]
- Tested with: undergraduates learning word pairs and prose in the laboratory; the classroom quizzing studies span school and university; the multiple-choice findings come to the wiki second-hand through one review chapter (Karpicke 2017).
- Not settled: whether an explanation beats the bare correct answer after an error; no claim here compares elaborated with answer-only feedback.
When should feedback come?¶
- Default: timing is not a strong lever in the evidence the wiki holds: in the one direct comparison here, immediate and delayed feedback after multiple-choice tests were equally effective at reducing lure intrusions — Providing feedback after initial multiple-choice tests cut lure intrusions on final short-answer tests roughly in half, with immediate and delayed feedback equally effective [~M]
- Changes when: second-language vocabulary is learned incidentally from reading → one study, reported second-hand in a review, found delayed feedback better than immediate for retention — Incidental vocabulary gains show retention problems, with losses over time moderated by feedback timing and topic interest [~W]
- Tested with: university students on laboratory retrieval tasks; L2 readers in one study cited by a review.
- Not settled: no wiki claim with its own evidence tests immediate against delayed feedback for procedural skills, clinical simulation or school mathematics; the timing advice in several claims' Discussion sections is not backed by their evidence entries.
Who should give it: teacher, peer, software or the learner?¶
- Default: peer and computer feedback are real options, not second-best. Structured peer feedback on university writing gave g = 0.91 against no feedback and did not differ reliably from teacher feedback; peer assessment across school and university levels gave g = 0.31 against no assessment and g = 0.28 against teacher assessment; feedback in computer-assisted language learning averaged g = 0.56 over 21 studies — Peer Feedback Improves Writing [+M], Peer assessment improves academic performance by a small average amount relative to no assessment or to teacher assessment [+M], Feedback in CALL has a significant moderate positive effect on student language learning outcomes (RE g = 0.56) across 21 studies [+M]
- Changes when: peers are the source → stronger writers gave more helpful comments, though not more accurate diagnoses; with a well-specified rubric, peer rating accuracy did not differ between higher- and lower-performing schools; and a rubric changed what peer reviewers commented on — In two secondary-school writing studies, peer reviewers' own performance predicted how helpful their feedback was but not how accurately it identified problems [~M], In two online comparisons, a rubric changed what peer reviewers commented on (one study) and helped peer raters place writing in quality tiers, most for complex criteria (the other); the first study says the rubric restrained reflective thinking [+W]
- Changes when: software is the source → marking an answer wrong without saying why was judged not beneficial by every participant in a small simulation pilot; the CALL effect also varied with educational level, provider, mother tongue, context, subject and target language — Students find computer-based simulation feedback insufficient because incorrect responses are marked without explanation [-W], Six of thirteen coded moderators significantly influence the effect of CALL feedback: educational level, intervention provider, mother tongue, research context, subject domain, and target language [~M]
- Changes when: learners assess themselves → formative assessment that supported student self-assessment gave d = .61 against .29 overall in US K-12, but untrained self-assessment is inaccurate, and studying modelled examples of self-assessment made it more accurate — Assessment for learning improves achievement [+M], Self Assessment Accuracy Is Low Without Training [~M]
- Tested with: higher-education writing (peer feedback), primary to tertiary students across subjects (peer assessment), language learners in CALL studies, US K-12 classrooms (formative assessment), secondary students (self-assessment training), 10 speech-language pathology graduate students (the simulation pilot).
- Not settled: whether automated feedback in a self-paced mobile app matches a teacher's; the CALL meta-analysis does not say, and its moderators name who delivered it, not which is better.
For second-language errors, which kind of correction?¶
- Default: correct errors; oral corrective feedback in 15 classroom studies had d = 0.74 against controls, and the effect held on delayed tests (0.63 immediate, 0.84 delayed) — Corrective feedback on second-language errors has a significant effect on L2 development that is maintained on delayed tests [+S]
- Changes when: choosing between recasts and prompts → in classrooms, prompts (d = 0.83) and explicit correction (d = 0.84) had larger effects than recasts (d = 0.53), though all three worked — The literature is supportive of recasts, but recast effectiveness varies with the language area and learner knowledge, focus, and aptitudes [~M]
- Changes when: the outcome is delayed rather than immediate → in one Iranian EFL study metalinguistic feedback beat recasts on the immediate test but not ten days later, because the metalinguistic group declined; a meta-analysis found implicit feedback better maintained than explicit, without saying how these two types were classed — Metalinguistic feedback outperforms recasts on the immediate post-test, but the two do not differ significantly on the delayed post-test [~M], Metalinguistic feedback's gains on wh-question performance decline over a ten-day delay while recast gains hold, leaving the two no different [~W]
- Tested with: classroom learners, mostly children and adolescents: adults (mean age above 23) were 140 of 827 learners in the classroom meta-analysis, and younger learners benefited more; one Iranian EFL wh-question study. Population-specific: for adult self-paced learners this evidence is the least representative.
- Not settled: corrective feedback delivered by an app rather than a teacher or interlocutor; treatment length (the two meta-analyses disagree).
Should learners choose when they get feedback?¶
- Changes when: learners may request feedback → letting them decide after a trial whether to see knowledge of results improved 24-hour retention over decisions made before the trial and over yoked controls; deciding before the trial gave no benefit — Learners who could decide after a trial whether to receive knowledge of results (Self-After, Self-Both) were more accurate on a 24-h motor retention test than learners who decided before the trial (Self-Before) and than their yoked counterparts. [+W], Having control over the knowledge-of-results decision only before a motor trial (Self-Before) produced no statistically significant retention or transfer benefit over a yoked group receiving the same KR schedule. [~W]
- Tested with: one laboratory study of university volunteers on a slider-positioning motor task. Population-specific: nothing here tests learner-requested feedback in cognitive, language or clinical tasks.
- Not settled: a default. No claim here sets how often feedback should be given, or whether reducing its frequency helps retention.
Is feedback the active ingredient, or the practice around it?¶
- Default: in simulation, pair feedback with repeated practice to a standard; simulation with deliberate practice (repetition, mastery standards, immediate feedback) beat traditional clinical education by d = 0.71 across 14 studies — Simulation Based Education With Deliberate Practice Improves Clinical Outcomes [+M]
- Changes when: asking whether feedback itself drives the gain → across 609 health-professions simulation studies, feedback, repetition and mastery learning showed no consistent significant interaction with the simulation effect; and in one small tutoring study, tutors told to withhold feedback and explanations and only prompt produced as much learning — Simulation Based Education Improves Outcomes [~M], Human tutoring has a medium effect over no tutoring, and in one small study students learned as well when tutors only prompted them as when tutors also explained and gave feedback [~W]
- Tested with: physicians, nurses and other health professionals on procedural skills (simulation); 8th-graders with college tutors, 11 dyads per study (tutoring).
- Not settled: debriefing as a feedback format; no claim here isolates it, and none tests medication-safety scenarios specifically.
The evidence by question¶
- What should the feedback after a check say, and who or what should give it? Feedback improves learning by a medium amount on average, but its effect varies widely with the information it carries, and a substantial share of feedback effects are negative [+S]
- Does confirming or correcting each response add to learning from making the response? Feedback Enhances Retrieval Practice [+S]
- Does corrective feedback on second-language errors improve accuracy, and which form of correction (recast, prompt, metalinguistic clue, explicit correction with or without a rule) works for which target and learner? Corrective feedback on second-language errors has a significant effect on L2 development that is maintained on delayed tests [+S]
- Can peers give the feedback, and how does peer feedback compare with none and with a teacher's? Peer assessment improves academic performance by a small average amount relative to no assessment or to teacher assessment [+M]
Related Elements¶
- Formative Assessment — the assessment structure that generates feedback while learning is still underway
- Coaching — the delivery mode in which feedback is embedded in ongoing, situated guidance
- Scaffolding — feedback functions as a temporary support that should fade with competence
- Practice — feedback without practice opportunities to apply it has little effect
- Assessment — the formal structure through which much feedback is delivered
Patterns That Use This Element¶
- Formative Assessment — feedback is the central mechanism of the formative cycle
- Competency-Based Learning — feedback against explicit competencies drives re-attempt and progression decisions
- Cognitive Apprenticeship — the coaching phase, where the expert diagnoses and corrects learner performance
Examples¶
- WAT-C classroom writing analytics platform — open-source NLP feedback on persuasive and source-based essays
Action-Oriented Feedback — Comments phrased as a next action ("add a counterexample here") rather than a judgment ("weak argument"), making the revision path explicit.
5-Minute Writing Conferences — Brief one-on-one conferences delivering targeted feedback on a single writing focus, keeping the loop short and actionable.
Khan Academy — Immediate item-level feedback on practice exercises with hints and linked explanations, allowing retry until mastery.
Draftback / Google Docs commenting — Inline, asynchronous teacher comments on drafts before final submission, preserving the revision cycle. - Provide timely automated feedback through an AI chatbot during EFL writing practice - Pair form-focused instruction with prompts that push learners to self-repair rather than recasts
Key Sources¶
- Hattie, J., & Timperley, H. (2007). The power of feedback. Review of Educational Research, 77(1), 81–112. doi:10.3102/003465430298487
- Shute, V. J. (2008). Focus on formative feedback. Review of Educational Research, 78(1), 153–189. doi:10.3102/0034654307313795
- Kluger, A. N., & DeNisi, A. (1996). The effects of feedback interventions on performance: A historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological Bulletin, 119(2), 254–284. doi:10.1037/0033-2909.119.2.254
- Black, P., & Wiliam, D. (1998). Assessment and classroom learning. Assessment in Education: Principles, Policy & Practice, 5(1), 7–74. doi:10.1080/0969595980050102
Page metadata
| Field | Value |
|---|---|
type |
element |
id |
feedback |
title |
Feedback |
description |
Learners receive input from instructors or peers to refine their work. |
status |
review |
canonical |
True |
generated |
by claude/unspecified on 2026-10-21 |
trust |
unverified — no one has confirmed this page's content |
Frontmatter is the page's machine-readable record — see the Schema & Guide.