Truace
Subscribe
70/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Meta-analysis of 33 studies finds AI tools tied to large gains for English learners

Study title: The cognitive-affective impact of generative AI on EFL/ESL learning: A meta-analysis

Evidence type
Systematic review or meta-analysis
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2820cite or check it

Study title: The cognitive-affective impact of generative AI on EFL/ESL learning: A meta-analysis

A systematic review and meta-analysis pooled 33 empirical studies, plus 6 studies summarised narratively, on how generative AI affects English as a foreign or second language learning. Covering work from November 2022 to March 2026, it found a large overall effect (Hedges' g = 1.253, 95% CI 0.954 to 1.542).

Read the study abstract

Since ChatGPT’s public release in November 2022, generative artificial intelligence (GenAI) has rapidly transformed English as a Foreign/Second Language (EFL/ESL) education, yet its precise impact on learning outcomes remains subject to debate. This systematic review and meta-analysis synthesizes evidence from 33 pooled empirical studies and 6 narrative-synthesis studies examining GenAI’s effects on language learning outcomes from November 2022 to March 2026. Following PRISMA 2020 guidelines, we searched multipl…

The gain

A meta-analysis of 33 pooled studies found generative AI use in English as a foreign or second language learning was linked to a large overall effect (Hedges' g = 1.253), with language performance at g = 1.197.

In brief

Language performance outcomes showed g = 1.197, and affective outcomes such as attitudes and feelings about learning showed g = 1.3347. Self-regulation outcomes had a larger estimate of g = 1.400, but it was not statistically significant (p = .187).

Results varied widely between studies, with heterogeneity (I²) between 88.4% and 99.1% across outcome domains. Research design and intervention duration were significant moderators, and experimental designs produced the strongest effects. Evidence certainty was rated moderate under GRADE.

Main points

  1. The review followed PRISMA 2020 guidelines and searched multiple databases for studies published from November 2022 to March 2026.
  2. It pooled 33 empirical studies and treated 6 more through narrative synthesis.
  3. Overall effect: Hedges' g = 1.253 (95% CI 0.954 to 1.542), with I² = 94.4%.
  4. Language performance g = 1.197 (I² = 88.4%); affective outcomes g = 1.3347 (I² = 98.3%).
  5. Self-regulation estimate g = 1.400 was large but not significant (p = .187, I² = 99.1%).
  6. Experimental designs produced the strongest effects, and research design and intervention duration were significant moderators.
  7. 87.9% of the literature is at the tertiary level; a tertiary-only sensitivity analysis gave consistent pooled estimates.
  8. Egger's test found no evidence of publication bias in the corrected dataset (p = .49).

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.

  • Overall language learning outcomes in EFL/ESL learners using generative AI

    Hedges' g 1.253 (95% CI 0.954 to 1.542)

    33 studies · search to March 2026

    The review’s own words
    Results revealed a large overall effect (Hedges’ g = 1.253, 95% CI [0.954, 1.542], I² = 94.4%).

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 applied a pre-specified effect-size verification rule to screen for reporting anomalies, then calculated random-effects models for three outcome domains: language performance, affective outcomes and self-regulation. The review covers evidence since ChatGPT's public release in November 2022.

The pooled effects were large, but the studies differed widely. Heterogeneity reached 94.4% overall and 99.1% for self-regulation, so the average figures hide wide variation between studies and settings.

The authors say the findings support generative AI's effectiveness in EFL/ESL learning, while stressing implementation design, the context-sensitivity of affective outcomes and the need for sustained metacognitive scaffolding to promote self-regulation. Evidence certainty was rated moderate using GRADE criteria.

Sources

1 on file
  1. Language Learning & Technologydoi:10.64152/10125/73714
    Peer-reviewed2026-01-01

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 33 studies finds AI tools tied to large gains for English learnershttps://truvace.com/item/the-cognitive-affective-impact-of-generative-ai-on-efl-esl-learning-a-meta-analy

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

    The gain A meta-analysis of 33 pooled studies found generative AI use in English as a foreign or second language learning was linked to a large overall effect (Hedges' g = 1.253), with language performance at g = 1.197.

    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 Language Learning & Technology.

    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 literature is overwhelmingly concentrated at the tertiary level (87.9%), and heterogeneity across studies was very high (I² up to 99.1%), with the self-regulation estimate not significant.

    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