Truace
Subscribe
70/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Students got moderate learning gains from AI feedback in a review of 32 Chinese studies

Study title: Meta-Analysis of the Impact of Artificial Intelligence-Based Teaching Feedback on Learning Outcomes

Evidence type
Systematic review or meta-analysis
Sample
n = 12,486
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2802cite or check it

Study title: Meta-Analysis of the Impact of Artificial Intelligence-Based Teaching Feedback on Learning Outcomes

A meta-analysis pooled 32 empirical studies published in China between 2016 and 2024, covering 12,486 students. It compared AI teaching feedback with traditional feedback and found a pooled effect of Hedges' g=0.51 (95% CI 0.42 to 0.60) on learning outcomes, which the author calls a moderate positive effect.

Read the study abstract

This study is grounded in the practice of educational digital transformation, and was designed in strict accordance with the methodological norms of meta-analysis. By systematically sorting out the inconsistent research conclusions on artificial intelligence teaching feedback, we systematically searched and screened the China National Knowledge Infrastructure (CNKI) and Wanfang Database, and formulated strict literature inclusion and exclusion criteria, we ultimately selected 32 eligible empirical studies publis…

The gain

A meta-analysis of 32 Chinese studies covering 12,486 students found AI teaching feedback raised learning outcomes over traditional feedback, with a pooled Hedges' g of 0.51, a moderate effect.

In brief

The size of the gain varied with school stage, feedback type and subject. Primary school pupils benefited most (g=0.62), followed by secondary students (g=0.49) and university students (g=0.38). Immediate and personalized feedback outperformed delayed and standardized feedback.

Heterogeneity between studies was high (I²=83.45%), and the included interventions mostly lasted 1 to 6 months, so lasting effects remain unmeasured.

Main points

  1. The analysis included 32 independent studies from 19 Chinese provinces and cities, drawn from the CNKI and Wanfang databases, with 12,486 participants in total.
  2. Pooled effect on learning outcomes was Hedges' g=0.51 (95% CI 0.42 to 0.60) under a random-effects model.
  3. By stage: primary school g=0.62 (11 studies), secondary g=0.49 (10 studies), university g=0.38 (11 studies).
  4. Immediate feedback (19 studies) gave g=0.58 against g=0.32 for delayed feedback (13 studies).
  5. Personalized feedback (17 studies) gave g=0.64 against g=0.35 for standardized feedback (15 studies).
  6. By subject: language g=0.57 (12 studies), mathematics and science g=0.48 (11 studies), skill-based subjects g=0.45 (9 studies).
  7. Publication-bias checks found no significant bias: Egger's test p=0.217 and a fail-safe N of 3,824.

What the review found when it combined the studies

These are the review’s own pooled results, copied from its abstract. Truvace did not recalculate them.

  • Learning outcomes with immediate AI teaching feedback

    Hedges' g 0.58 (95% CI 0.47 to 0.69)

    19 studies

    The review’s own words
    g=0.58, 95% confidence interval [0.47, 0.69], p<0.001,

  • Student learning outcomes with AI teaching feedback

    Hedges' g 0.51 (95% CI 0.42 to 0.60)

    32 studies · 12,486 people

    The review’s own words
    The results revealed that artificial intelligence teaching feedback had a pooled effect size of Hedges' g=0.51 (95% CI [0.42, 0.60], Z=11.24, p<0.001) on learning outcomes, representing

  • Learning outcomes of middle school students with AI teaching feedback

    Hedges' g 0.49 (95% CI 0.35 to 0.63)

    The review’s own words
    with a pooled effect size of Hedges' g=0.49 and a 95% confidence interval [0.35, 0.63], p<0.001, indicating a moderate positive effect;

  • Maths and science learning outcomes with AI teaching feedback

    Hedges' g 0.48 (95% CI 0.36 to 0.60)

    The review’s own words
    The pooled effect size Hedges' g=0.48, 95% confidence interval [0.36, 0.60], p<0.001,

  • Learning outcomes of primary school students with AI teaching feedback

    Hedges' g 0.62 (95% CI 0.48 to 0.76)

    11 studies

    The review’s own words
    pooled effect size of Hedges' g=0.62, a 95% confidence interval [0.48, 0.76], and p<0.001.

Code checks every number above against the review’s text before it is shown. The short labels are an AI model’s reading of that text, so the review’s own sentence sits under each one. A pooled figure is only as strong as the studies behind it. How Truvace uses pooled results

The full story

The author followed PRISMA reporting standards and searched two Chinese databases, CNKI and Wanfang. Included studies needed an AI-feedback group and a traditional-feedback control group, plus data for calculating effect sizes. Two researchers screened studies and extracted data independently, and the analysis used CMA 3.0 software.

Between-study heterogeneity was high (Q=187.23, df=31, I²=83.45%), so a random-effects model was used. The funnel plot was roughly symmetrical, Egger's test gave p=0.217, and the fail-safe N of 3,824 far exceeded the 170 threshold.

Group differences were statistically significant for stage (Q=8.72, p<0.05) and subject (Q=6.34, p<0.05). The author says language learning benefits from frequent instant feedback, while mathematics and skills subjects need deeper feedback that current AI tools struggle to give.

The author recommends educators prioritize real-time, personalized AI feedback matched to students' stage and subject, and calls for further work on adaptive AI feedback technologies.

Sources

1 on file
  1. International Journal of Education and Humanitiesdoi:10.54097/zq9yk850
    Peer-reviewed2026-04-20

Talk it over

Four questions to start onWho gets this benefit, and who was left out of the study or story?

Start with a question

Students got moderate learning gains from AI feedback in a review of 32 Chinese studieshttps://truvace.com/item/meta-analysis-of-the-impact-of-artificial-intelligence-based-teaching-feedback-o

  1. Who gets this benefit, and who was left out of the study or story?

    The gain A meta-analysis of 32 Chinese studies covering 12,486 students found AI teaching feedback raised learning outcomes over traditional feedback, with a pooled Hedges' g of 0.51, a moderate effect.

    Answer in the thread

  2. How do we know? What can this kind of evidence show, and what can’t it?

    The evidence This rests on a systematic review or meta-analysis with a sample of 12,486, from International Journal of Education and Humanities.

    Answer in the thread

  3. What does it leave open? What else would you want to know before acting on it?

    The record’s own caveat The included studies mostly ran 1 to 6 months, so the review could not examine long-term effects of AI teaching feedback; it also drew only on studies published in China.

    Answer in the thread

  4. Have you seen AI used this way where you study, work or live? Did it go the same way?

    Answer in the thread

The questions are the same on every record. The details under them are taken from this record as written.

What readers say

More from Education