Truace
Subscribe
70/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Meta-analysis of 53 studies finds AI-assisted teaching beats non-AI methods on grades and writing

Study title: Generative AI technologies and educational outcomes: a comprehensive meta-analysis comparing traditional and AI-driven approaches

Evidence type
Systematic review or meta-analysis
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2813cite or check it

Study title: Generative AI technologies and educational outcomes: a comprehensive meta-analysis comparing traditional and AI-driven approaches

A meta-analysis following the PRISMA protocol pooled 53 studies comparing generative AI with non-AI teaching. Students taught with generative AI scored higher on academic achievement (g=0.40), higher-order thinking (g=0.72), learning motivation (g=0.81) and writing skills (g=0.76).

Read the study abstract

With the exponentially rapid development of information technologies, generative artificial intelligence (GenAI) technologies have been widely applied to various fields, including education. However, few studies have synthesized the pooled effect of GenAI on educational outcomes, which is an important research gap this study aims to address. Through the PRISMA-based meta-analysis and systematic review, this study concludes a central finding: GenAI technologies may generally outperform traditional and non-GenAI e…

The gain

A meta-analysis of 53 studies found generative AI-assisted education produced higher academic achievement (g=0.40), higher-order thinking (g=0.72) and writing skills (g=0.76) than non-AI approaches.

In brief

AI-generated feedback showed the largest pooled effect (g=1.27) against non-AI feedback. Game-assisted generative AI did not beat non-game-assisted AI (g=0.24, confidence interval from -0.40 to 0.88).

Results differed by country. Gains were significant in China (g=0.71) and Pakistan (g=0.75), but not in Korea or Turkey (g=0.01). Effects were significant at both university (g=0.70) and secondary level (g=0.80).

Main points

  1. The search covered Web of Science, Scopus, Springer, Wiley and ScienceDirect, plus dissertations and conference papers, and returned 1,096 records; 53 studies met the criteria.
  2. Of the 683 full texts assessed, 630 were excluded, mostly for missing key data (267) or unreliable designs (205).
  3. By the authors' quality scoring, 28 of the 53 studies were high quality, 19 medium and 6 low.
  4. Pooled effects: academic achievement g=0.40, higher-order thinking g=0.72, writing g=0.76, motivation g=0.81.
  5. AI feedback produced g=1.27 versus non-AI feedback, while game-assisted AI showed no significant difference.
  6. In Korea the interval for the effect ran from -0.41 to 3.78, and in Turkey the effect was essentially nil (g=0.01).
  7. University students (g=0.70) and secondary students (g=0.80) both showed significant gains.

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.

  • Academic achievement with generative AI-assisted education compared with non-generative-AI approaches

    Hedges' g 0.40 (95% CI 0.08 to 0.71)

    53 studies

    The review’s own words
    GenAI-assisted education may cause significantly higher academic achievements (g=0.40, small-to-moderate effect; 95%CI=0.08~0.71), higher-order thinking abilities (g=0.72; 95%CI=0.30~1.13, moderate-to-large effect), learning motivation (g=0.81; 95%CI=0.20~1.41, large effect), and writing skills (g=0.76; 95%CI=0.18~1.35, moderate-to-large effect) than the non-

  • Educational outcomes with game-assisted generative AI compared with non-game-assisted generative AI

    Hedges' g 0.24 (95% CI -0.40 to 0.88)

    53 studies

    The review’s own words
    It is demonstrated that GenAI feedback may exert a significant influence on educational outcomes (g=1.27; 95%CI=0.49~2.05), while game-assisted GenAI may not cause significant differences in educational outcomes (g=0.24; 95%CI=-0.40~0.88).

  • Educational outcomes with generative AI feedback compared with non-generative-AI feedback

    Hedges' g 1.27 (95% CI 0.49 to 2.05)

    53 studies

    The review’s own words
    It is demonstrated that GenAI feedback may exert a significant influence on educational outcomes (g=1.27; 95%CI=0.49~2.05), while game-assisted GenAI may not cause significant differences in educational outcomes (g=0.24; 95%CI=-0.40~0.88).

  • Writing skills with generative AI-assisted education compared with non-generative-AI approaches

    Hedges' g 0.76 (95% CI 0.18 to 1.35)

    53 studies

    The review’s own words
    GenAI-assisted education may cause significantly higher academic achievements (g=0.40, small-to-moderate effect; 95%CI=0.08~0.71), higher-order thinking abilities (g=0.72; 95%CI=0.30~1.13, moderate-to-large effect), learning motivation (g=0.81; 95%CI=0.20~1.41, large effect), and writing skills (g=0.76; 95%CI=0.18~1.35, moderate-to-large effect) than the non-

  • Learning motivation with generative AI-assisted education compared with non-generative-AI approaches

    Hedges' g 0.81 (95% CI 0.20 to 1.41)

    53 studies

    The review’s own words
    GenAI-assisted education may cause significantly higher academic achievements (g=0.40, small-to-moderate effect; 95%CI=0.08~0.71), higher-order thinking abilities (g=0.72; 95%CI=0.30~1.13, moderate-to-large effect), learning motivation (g=0.81; 95%CI=0.20~1.41, large effect), and writing skills (g=0.76; 95%CI=0.18~1.35, moderate-to-large effect) than the non-

  • Educational outcomes with generative AI in Pakistan compared with non-generative-AI approaches

    Hedges' g 0.75 (95% CI 0.58 to 0.92)

    53 studies

    The review’s own words
    As shown in Figure 5, in China (g=0.71; 95%CI=0.5~0.92) and Pakistan (g=0.75; 95%CI=0.58~0.92), GenAI technologies cause significantly higher educational outcomes than the non-GenAI technologies.

  • Educational outcomes with generative AI in China compared with non-generative-AI approaches

    Hedges' g 0.71 (95% CI 0.5 to 0.92)

    53 studies

    The review’s own words
    As shown in Figure 5, in China (g=0.71; 95%CI=0.5~0.92) and Pakistan (g=0.75; 95%CI=0.58~0.92), GenAI technologies cause significantly higher educational outcomes than the non-GenAI technologies.

  • Higher-order thinking abilities with generative AI-assisted education compared with non-generative-AI approaches

    Hedges' g 0.72 (95% CI 0.30 to 1.13)

    53 studies

    The review’s own words
    GenAI-assisted education may cause significantly higher academic achievements (g=0.40, small-to-moderate effect; 95%CI=0.08~0.71), higher-order thinking abilities (g=0.72; 95%CI=0.30~1.13, moderate-to-large effect), learning motivation (g=0.81; 95%CI=0.20~1.41, large effect), and writing skills (g=0.76; 95%CI=0.18~1.35, moderate-to-large effect) than the non-

  • Educational outcomes with generative AI at secondary level compared with non-generative-AI approaches

    Hedges' g .80 (95% CI 0.15 to 1.44)

    53 studies

    The review’s own words
    At both university (g=.70; 95%CI=0.52~0.88) and secondary (g=.80; 95%CI=0.15~1.44) educational levels, GenAI technologies cause significant differences in educational outcomes (Figure 6).

  • Educational outcomes with generative AI at university level compared with non-generative-AI approaches

    Hedges' g .70 (95% CI 0.52 to 0.88)

    53 studies

    The review’s own words
    At both university (g=.70; 95%CI=0.52~0.88) and secondary (g=.80; 95%CI=0.15~1.44) educational levels, GenAI technologies cause significant differences in educational outcomes (Figure 6).

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 authors searched five databases in March 2025 for experimental, quasi-experimental or survey studies with a control group. Two independent raters screened records, with a third rater resolving disagreements. Effects were expressed as Hedges' g and pooled with a random-effects model after a Galbraith plot showed heterogeneity.

The included studies cover ChatGPT use in computer programming, physics, English writing, medical education and music. Academic achievement was measured mainly through standardized exams, adapted tests and assignments; writing was scored for cohesion, accuracy and content innovation.

Country results split sharply. The authors attribute the null results in Korea and Turkey to cultural preference for teacher-led instruction, curricular misalignment and technical limitations, whereas China and Pakistan used GenAI to extend resources to remote areas. These explanations come from the cited literature rather than being tested directly.

The authors suggest future work examine students with disabilities and personalised teaching strategies.

Sources

1 on file
  1. Humanities and Social Sciences Communicationsdoi:10.1057/s41599-026-06903-y
    Peer-reviewed2026-03-12

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

Meta-analysis of 53 studies finds AI-assisted teaching beats non-AI methods on grades and writinghttps://truvace.com/item/generative-ai-technologies-and-educational-outcomes-a-comprehensive-meta-analysi

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

    The gain A meta-analysis of 53 studies found generative AI-assisted education produced higher academic achievement (g=0.40), higher-order thinking (g=0.72) and writing skills (g=0.76) than non-AI approaches.

    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, from Humanities and Social Sciences Communications.

    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 Effect sizes were significantly heterogeneous across studies (I²>50%), and several included studies were rated medium or low quality, so pooled estimates vary widely and have wide confidence intervals.

    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