TAs adapt rather than copy AI feedback drafts, with moderate draft–final overlap (ROUGE-L F1 = 0.60)¶
Claim · All claims
Evidence · 1 study · causalr2·q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person
Subclaims¶
Reading the evidence codes
q is the design tier, i the impact, n the sample, and the kind with its r
the rigour judged within that kind: what each code means.
This claim: no standardized effect is recorded in observations/ for it yet, so no benchmark reading is possible.
q2 i? Average ROUGE-L F1 similarity between draft and final feedback was 0.60 at course level, suggesting TAs use drafts as starting points and adapt them. → Mahinpei 2026
Evidence¶
Mahinpei 2026¶
Mahinpei, R., Dean, V., Fong, R., Liu, L. T., & Ribeiro, M. H. (2026). AI Assistance for Discretionary Work: Increasing Feedback Provision in Higher Education. arXiv. https://arxiv.org/abs/2606.03095 doi:10.48550/arxiv.2606.03095
q2 · i? · causal · r2
Descriptive treatment-only analysis of similarity scores (N = 432) computed via ROUGE-L F1 whenever both draft and final feedback were available. The article reports "moderate lexical overlap" and interprets drafts as intermediate artifacts preserving TA judgment.
"We find moderate lexical overlap between the feedback draft and the final feedback from TAs, with an average ROUGE-L F1 score of 0.60 at the course level. This suggests that TAs often use the feedback draft as a starting point, adapting it"
Discussion¶
Related Claims¶
- AI-assisted feedback drafts increase discretionary feedback provision by TAs by 10.81 percentage points — related
- AI-assisted feedback drafts do not reduce TAs' time spent per character of feedback — related
- Draft usefulness varies by grading context: higher for submissions with mistakes and applied questions — related
- In the Enacted Feedback workflow most resources moved from AI feedback to suggestion selection, while optional AI assistance was rarely entered — related
- AI-assisted feedback drafts increase total feedback length by 39.79 characters per submission — related
- AI-assisted feedback does not reduce student-perceived usefulness of rated feedback — related
Page metadata
| Field | Value |
|---|---|
type |
claim |
id |
tas-adapt-rather-than-copy-ai-drafts |
title |
TAs adapt rather than copy AI feedback drafts, with moderate draft–final overlap (ROUGE-L F1 = 0.60) |
description |
TAs adapt rather than copy AI feedback drafts, with moderate draft–final overlap (ROUGE-L F1 = 0.60) |
status |
review |
evidence_strength |
moderate |
newest_evidence |
2026 |
generated |
by process:wiki-ingest on 2026-10-10 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — mahinpei-2026 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.