Differences between human and ChatGPT feedback quality were small, with means within one point and partial eta-squared ranging from .01 to .12¶
Claim · All claims
Evidence · 1 study · causalr2·q3
Currency · newest study 2023 · not yet checked against its sources · not reviewed by a person · recheck by 2026: tool and model accuracy changes quickly
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.
q3 i? Mean differences between human and AI feedback ranged from -.24 (favoring AI) to .83 (favoring humans), and partial eta-squared values ranged from .01 to .12. → Steiss 2023
Evidence¶
Steiss 2023¶
Steiss, Jacob; Tate, Tamara; Graham, Steve; Cruz, Jazmin; Hebert, Michael; Wang, Jiali; Moon, Youngsun; Tseng, Waverly; Warschauer, Mark. (2023). Comparing the Quality of Human and ChatGPT Feedback on Students' Writing. https://eric.ed.gov/?q=Comparing+the+Quality+of+Human+and+ChatGPT+Feedback+on+Students+Writing
q3 · i? · causal · r2
Descriptive and ANOVA analysis of coded feedback ratings (n = 198) found "all small" partial eta-squared values from .01 to .12; ChatGPT means ranged from 3.09 to 4.02 across dimensions.
"Partial eta-squared effect sizes for variance explained by the difference in rater were all small, ranging from .01 (for criteria-based feedback) to .12 (for prioritization of essential features), indicating that the difference in raters explained very little of the differences in feedback ratings."
Discussion¶
Related Claims¶
- Trained human evaluators provided higher-quality formative feedback than ChatGPT on four of five components, while ChatGPT outperformed humans on criteria-based feedback — related
- Feedback quality did not differ by students' language status for either human evaluators or ChatGPT — related
- LLM-generated 7C collaboration assessment scores fall within the range of human expert variability across ten discussions — related
- Students rate LLM-generated response-level feedback favourably, with clarity and perceived accuracy highest and actionability and usefulness lowest — related
- Accurate LLM solution diagnosis does not reliably produce pedagogically actionable feedback, with lowest rubric scores on error identification and actionability — related
- 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 — related
- Students rate human-generated feedback as superior to AI-generated feedback in quality, helpfulness, and clarity, and as more likely to be acted upon — related
Page metadata
| Field | Value |
|---|---|
type |
claim |
id |
small-effect-sizes-human-ai-feedback-differences |
title |
Differences between human and ChatGPT feedback quality were small, with means within one point and partial eta-squared ranging from .01 to .12 |
description |
Differences between human and ChatGPT feedback quality were small, with means within one point and partial eta-squared ranging from .01 to .12 |
status |
review |
evidence_strength |
moderate |
newest_evidence |
2023 |
recheck_by |
2026 |
generated |
by process:wiki-ingest on 2026-10-15 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — steiss-2023 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.