Moderate evidence
More
Language learners wrote better with AI, a review of 16 studies shows
Study title: How effective is generative artificial intelligence (GenAI) on task-based language teaching (TBLT) for writing: A meta-analysis
- Evidence type
- Systematic review or meta-analysis
- Sources
- 1 sourcePeer-reviewed
- Published
- 2 min read
- Truvace record
- TRV-2026-3089cite or check it
Study title: How effective is generative artificial intelligence (GenAI) on task-based language teaching (TBLT) for writing: A meta-analysis
A meta-analysis by researchers at The Education University of Hong Kong, Yildiz Technical University and The Chinese University of Hong Kong pooled 16 studies on generative AI in task-based language teaching for writing. Students who learned with generative AI outperformed those who learned without it, with a pooled effect size of g = 0.76 (p < 0.05).
Read the study abstract
Generative artificial intelligence (GenAI) has influenced writing significantly, especially for second language (L2) learners; however, limited literature has systematically examined the effectiveness of GenAI on writing. Previous meta-analyses present inconsistent findings on the effectiveness of GenAI, and these meta-analyses lack exploration from an instructional and pedagogical perspective. The task-based language teaching (TBLT) approach has been widely applied in language education. This approach concerns…

The gain
In a meta-analysis of 16 studies published 2022 to 2025, students who learned task-based writing with generative AI outperformed those who learned without it (g = 0.76).
In brief
The authors searched Web of Science, Scopus and ProQuest for studies published between 2022 and 2025, following PRISMA procedures. They set out to resolve inconsistent findings in earlier meta-analyses, which they say did not examine the question from an instructional and pedagogical angle.
A meta-regression found that educational level, language proficiency and task planning moderated the effect, and that these factors worked jointly on writing performance. The authors offer suggestions for teachers and future research.
Main points
- The meta-analysis pooled 16 studies on generative AI in task-based language teaching for writing, published between 2022 and 2025.
- Learners using generative AI outperformed those who did not, with an effect size of g = 0.76 (p < 0.05).
- Studies were collected from Web of Science, Scopus and ProQuest following PRISMA.
- Meta-regression identified three moderators of the effect: educational level, language proficiency and task planning.
- The analysis used a Triadic Componential Framework covering task difficulty, task complexity and task conditions.
The full story
The work was presented as a poster at ASCILITE 2026 and written by Yan Li, Bünyamin Bavli and Thomas K.F. Chiu. It targets second-language learners, for whom the authors say generative AI has significantly influenced writing but systematic evidence is limited.
Task-based language teaching is widely used in language education and attends to task characteristics and learner factors. The authors organised their analysis around a Triadic Componential Framework: task difficulty, task complexity and task conditions, which map to learner, cognitive and interactional factors.
Beyond the headline effect, the meta-regression showed that the benefit varied with students' educational level, language proficiency and whether tasks involved planning, and that these factors interact. The authors draw practical suggestions for teachers from these moderators.
Sources
1 on file- ASCILITE Publicationsdoi:10.65106/apubs.2026.3289
Talk it over
The meta-analysis of 16 studies found learners using generative AI outperformed those who did not, with g = 0.76. Which kinds of writing progress would you want that number to reflect?
Language learners wrote better with AI, a review of 16 studies showshttps://truvace.com/item/how-effective-is-generative-artificial-intelligence-genai-on-task-based-language
The meta-analysis of 16 studies found learners using generative AI outperformed those who did not, with g = 0.76. Which kinds of writing progress would you want that number to reflect?
Answer in the threadThe meta-regression found that educational level, language proficiency and task planning moderated the effect. Which learners might gain least, and how could that change what the average result means?
Answer in the threadThe authors say earlier meta-analyses gave inconsistent findings. What kind of result from a new study would make you doubt the g = 0.76 finding?
Answer in the thread
If you have more time
The studies were published between 2022 and 2025 and found through Web of Science, Scopus and ProQuest. What would you look up to see how long learners were followed after the tasks?
Answer in the threadThe authors organised their analysis around task difficulty, task complexity and task conditions. How would you use that framework to plan a writing task in a class you know?
Answer in the thread
ace
What readers say