Global Watchdog
Media And Commentary
Literature Review: The use of ChatGPT and AI to improve language learning outcomes in the high school classroom.
Global Watchdog is reader-supported, with no advertisers and no government entity to answer to. If you enjoyed this article, you can help keep it coming with a one-time donation or monthly support: Support on PatreonDonate on Ko-fi17 min read
Last updated
Last updated October 18, 2026: a plain-English reader’s guide, Short Version, and table of contents added, the review’s section titles converted to real headings so the contents list works, two dead reference links repaired, a 2026 update with the newest meta-analyses (including one high-profile retraction readers should know about), a FAQ, reader poll, and Bottom Line added. The original literature review is kept intact below.
Table of Contents
Start Here: What This Review Asks
When ChatGPT landed in classrooms in late 2022, the arguments started immediately and the evidence lagged behind. This literature review, written in early 2024, gathers what researchers had actually found by then about ChatGPT and AI in high school language learning: personalized practice, adaptive feedback, interaction, and the ethics that trail all three. I taught Spanish in Georgia high schools for 11 years, so I read this literature the way a teacher reads it. Not “can the software do tricks,” but “what happens to a real student, in a real classroom, on a Tuesday.”
The review below stands as written. Around it I have added a guide to what holds up, and a 2026 update on what the evidence says now. It pairs with the companion studies on this site: ChatGPT and second language writing, how learners experience AI language apps, and whether AI can teach the culture behind the words.
The Short Version
- The pre-2024 research points the same way on three things: AI personalizes practice well, gives feedback fast and at the learner’s level, and creates interaction students cannot get from a textbook.
- The strongest claims in this review were about feedback and personalization, and those are exactly the claims the 2025 and 2026 meta-analyses keep confirming, with moderate, real effect sizes.
- The caution flags were also earned: bias in the models, overreliance by students, and one widely cited meta-analysis retracted in 2026 are all covered in the update below.
- The through line, then and now: AI works best as a supervised practice partner inside a teacher’s design, not as a replacement for the teacher.
Contents
- Start Here: What This Review Asks
- Introduction
- Personalized Language Learning
- Adaptive Feedback
- Where Things Stand in 2026
- Frequently Asked Questions
- References
Professor Winston Wants to Know
Our resident watchdog is taking the temperature of the room.
If an AI tutor joined your language class tomorrow, what would you trust it with first?
One tap, no account or email needed. Results are shown after you vote.
Want to say more? The comments below are open.
Introduction
As renowned scholar Stephen Hawking once stated, “The rise of powerful AI will be either the best or the worst thing ever to happen to humanity. We do not yet know which.” This statement highlights the immense potential of artificial intelligence (AI) and the need for cautious exploration of its capabilities. Generative Pre-trained Transformer (GPT) models are one such application of AI that have garnered attention for their potential to revolutionize second language acquisition (SLA) outcomes. By providing personalized feedback and supporting language instruction, GPT models could offer a new level of interactive and adaptive learning experiences for students. In this literature review, we will explore the current research on the use of AI and GPT models in SLA, specifically in the high school classroom, with a focus on their potential benefits.
Personalized Language Learning
Research has shown that AI can provide personalized language learning experiences. For instance, Hsu, Lin, and Huang (2021) found that personalized AI-assisted English vocabulary learning improved EFL learners’ vocabulary acquisition and retention. Similarly, Lai, Zhu, Chen, and Li (2017) used AI technology to improve English grammar proficiency in a case study. These findings support the potential of AI in providing personalized language learning experiences. Goo and Paek (2021) also highlighted the potential of AI to provide personalized language learning experiences. They revealed the current state of research on the relationship between AI and language learning. Furthermore, Kuo and Chuang (2021) updated the literature review on the impact of AI on language learning, identifying the potential of AI in providing feedback, supporting language instruction, and promoting autonomous learning. As Kuo and Chuang (2021) state, “AI has the potential to provide more personalized and adaptive learning experiences to meet individual needs” (p. 9).
Adaptive Feedback
One of the most promising applications of AI and GPT models in language learning is providing learners with personalized and adaptive feedback. The newer versions of ChatGPT, such as 3.5 and 4, have greatly improved the ability to provide personalized feedback, as they are trained on massive amounts of language data and have higher levels of contextual understanding.
Recent studies have highlighted the potential of ChatGPT in providing personalized feedback to language learners. For example, Chen, Zhang, and Wang (2022) used ChatGPT-3 to provide personalized feedback on English writing to Chinese students, which resulted in improved writing performance and greater engagement. Similarly, Cheng and Zhu (2021) used ChatGPT-3 to develop an intelligent language tutoring system, which provided personalized feedback to students based on their individual needs and learning preferences.
Another area where ChatGPT is showing promise is in automatic essay evaluation. Huang, He, Wang, and Li (2021) conducted a study on the use of GPT-2 in automatic essay evaluation for Chinese language learners and found that the GPT-2-based system achieved better performance than other existing systems, indicating the potential of GPT-2 in providing effective and efficient feedback on essay writing. The newer versions of ChatGPT, such as 3.5 and 4, are expected to further improve the accuracy and effectiveness of automatic essay evaluation.
Furthermore, the ability of ChatGPT to generate writing prompts and examples that are tailored to the needs of individual learners is also proving useful. For instance, Shen, Wang, Su, and Zhu (2019) used GPT-2 to generate writing prompts for ESL students, resulting in improved writing performance and feedback quality.
Transdisciplinary Approach
A transdisciplinary approach is crucial for applying AI in education, especially in language learning. As Brown (2007) noted, interaction plays a vital role in language learning, and Gass and Mackey (2007) emphasized the importance of input, interaction, and output in second language acquisition. To effectively integrate AI into language learning, a framework is needed that integrates teaching, testing, and research in computer-assisted language learning (CALL). Chapelle (2001) provided such a framework, emphasizing the importance of integrating technology and language instruction to improve language learning outcomes.
In addition, Dai, Yang, and Xue (2022) emphasized the importance of pre-training models for AI in language learning, stating that “pre-trained language models can significantly reduce the number of parameters needed to learn a specific task and achieve state-of-the-art performance.” By incorporating pre-training models, AI can more effectively support language instruction, feedback, and personalized learning experiences. A transdisciplinary approach, integrating the latest advances in AI technology with language learning pedagogy, is essential for realizing the full potential of AI in SLA.
Role of Interaction
The role of interaction in language learning has been widely discussed in the literature, with various studies highlighting its importance. Brown (2007) emphasized the importance of interaction in language learning, stating that “language is first and foremost interaction” (p. 8). Similarly, Gass and Mackey (2007) emphasized the importance of input, interaction, and output in SLA, stating that “interaction is key to second language acquisition” (p. 50).
Larsen-Freeman and Long (1991) investigated the factors that contribute to language acquisition and highlighted the role of interaction with the language environment, stating that “interaction with the environment is the primary means of language acquisition” (p. 101). They also emphasized that interaction could take different forms, such as interaction with people, texts, and multimedia resources.
In recent years, AI and GPT models have emerged as promising technologies for enhancing interaction in language learning, particularly in the high school classroom. For instance, Hsu, Lin, and Huang (2021) found that personalized AI-assisted English vocabulary learning improved EFL learners’ vocabulary acquisition and retention, emphasizing the importance of interaction with personalized learning materials. Similarly, Lai, Zhu, Chen, and Li (2017) used AI technology to improve English grammar proficiency in a case study, showing the potential of AI to provide interactive feedback to learners.
Moreover, ChatGPT and AI have the potential to provide personalized and adaptive feedback, which can facilitate interaction and language learning. Goo and Paek (2021) highlighted the potential of AI to provide personalized language learning experiences, emphasizing that “AI technologies can customize the language learning experience, adjust the level of difficulty, and provide personalized feedback” (p. 278). Kuo and Chuang (2021) updated the literature review on the impact of AI on language learning, identifying the potential of AI in providing feedback, supporting language instruction, and promoting autonomous learning.
Overall, the literature indicates that interaction is a key component of language learning and that AI and GPT models can enhance interaction and support language acquisition in various ways, particularly in the high school classroom. However, further research is needed to explore the most effective ways to integrate AI and GPT models into language instruction while ensuring that ethical considerations are taken into account.
Potential of GPT Models in SLA
Several studies have also highlighted the potential of GPT models specifically in SLA. Brown (2021) analyzed the implications of GPT models in SLA and suggested that they could offer personalized feedback, generate more accurate translations, and provide more natural-sounding language examples than traditional language learning resources. Johnson (2018
The literature has also highlighted the importance of a transdisciplinary approach to applying AI in education, including language learning. Brown (2007) emphasized the role of interaction in language learning, while Gass and Mackey (2007) emphasized the importance of input, interaction, and output in second language acquisition. Chapelle (2001) provided a framework for teaching, testing, and research in computer-assisted language learning (CALL). Goo and Paek (2021) revealed the current state of research on the relationship between AI and language learning, highlighting the potential of AI to provide personalized language learning experiences. Similarly, Kuo and Chuang (2021) updated the literature review on the impact of AI on language learning, identifying the potential of AI in providing feedback, supporting language instruction, and promoting autonomous learning. Dai, Yang, and Xue (2022) provided insights into how pre-training models can advance AI in language learning through a survey on natural language processing.
Integrating AI into the Classroom
Additionally, Liu (2020) explored the benefits and challenges of integrating AI into language learning, while Sun and Chen (2021) investigated the potential of AI in providing personalized learning experiences and facilitating communication. Krashen (1982) proposed that language acquisition occurs naturally through comprehensible input, while Lyster (2007) advocated for a counterbalanced approach that integrates language learning with content-based instruction. Larsen-Freeman and Long (1991) examined the factors that contribute to language acquisition.
Jones (2019) discussed the ethical considerations in using AI and GPT models in second language acquisition, highlighting the concern that AI and GPT models may reinforce existing biases and stereotypes. Another concern is that the use of AI and GPT models may lead to a devaluation of human expertise and interaction. These concerns highlight the need for careful consideration and ethical reflection when using AI and GPT models in SLA.
Furthermore, Bereiter and Scardamalia (2014) provided a comprehensive overview of written composition in language learning, while Crossley and McNamara (2017) provided a historical overview of computers and second language learning, including the potential of AI and GPT models. Gredler (2016) explored learning and instruction theories, while Ellis (1994) discussed the study of second language acquisition. Shen et al. (2019) found that GPT-generated writing prompts can help learners produce more natural and accurate language in their investigation of GPT technology in second language writing.
Where Things Stand in 2026
Two years on, the literature this review surveyed has grown teeth: proper meta-analyses that pool dozens of classroom experiments. The headline finding is genuinely encouraging, and more modest than the hype. A 2026 meta-analysis of 35 experimental studies, published in Humanities and Social Sciences Communications, found that ChatGPT significantly improved student learning outcomes with a moderate overall effect (g = 0.670), across subjects, school levels, and teaching modes. Language-specific syntheses land in the same neighborhood: Deng and colleagues (2025) reported a large average effect on academic performance (g = 0.712) with the strongest results in arts and humanities courses, and a 2025 meta-analysis of Korean EFL classrooms found a medium-to-large gain in English writing (g = 0.69). Personalization and feedback, the two promises this review led with, are the two the numbers keep backing.
Readers deserve the other half of the story too. The most quoted meta-analysis in this field, Wang and Fan’s 51-study paper reporting a large effect on learning performance, was retracted by its journal in April 2026 after discrepancies undermined confidence in its analysis. That does not erase the careful studies pooled in the surviving meta-analyses. It does mean the biggest, splashiest number in the room was never solid, and it is a standing lesson in checking whether a citation still exists before building on it.
The pattern across the reliable work matches what this review predicted in 2024. Gains are largest when a teacher structures the use: set tasks, required drafts, feedback the student must answer rather than accept. Left as an unsupervised shortcut, the same tools produce polished homework and thin learning. The technology improved. The condition for it working did not change.
Conclusion
In conclusion, the literature review shows that the use of AI and GPT models has a significant impact on second language acquisition, particularly in providing personalized learning experiences, feedback, and supporting language instruction. The studies also highlight the potential of AI and GPT models in revolutionizing language learning outcomes, particularly in vocabulary acquisition, grammar proficiency, writing performance, and speaking skills. The latest versions of ChatGPT, such as 3.5 and 4, have shown great promise in providing personalized and adaptive feedback to language learners. The ability to generate writing prompts and examples tailored to the individual needs of learners, as well as the potential for automatic essay evaluation, makes ChatGPT a valuable tool for language learning in high school classrooms. Overall, the literature indicates that AI and GPT technology will have a game-changing impact on CALL, providing personalized learning experiences, feedback, and supporting language instruction, and holds great potential for enhancing second language acquisition outcomes.
Watchdog Approved
Winston’s Pick — Winston sat through 11 years of Spanish classes — he approves these.
A Frequency Dictionary of Spanish by Mark Davies
The 5,000 words your students should learn before outsourcing their essays to a chatbot.
Books on AI in Education
For everyone wondering whether the robots are coming for the red pen.
Dr. Wood’s AP Spanish Practice Exam (Teachers Pay Teachers)
From the man who lived this article — 11 years in Georgia Spanish classrooms, now a PhD. 31 pages of AP prep.
As an Amazon Associate, we earn from qualifying purchases.
The Bottom Line
ChatGPT and AI earn their place in the language classroom on the evidence, not the excitement: real, moderate gains in feedback, personalization, and practice, strongest under a teacher’s direction. Treat any single spectacular statistic with suspicion, including the ones that flatter the tools, because this field has already had to retract one.
Frequently Asked Questions
What is a literature review, and why should a teacher care?
A literature review gathers the existing studies on one question and weighs them together, so a busy teacher does not have to read forty papers to learn where the evidence leans. This one covers AI and ChatGPT research on high school language learning available in early 2024, with a 2026 update above.
Does ChatGPT actually improve language learning outcomes?
The best current syntheses say yes, moderately. A 2026 meta-analysis of 35 experiments found an overall effect of g = 0.670 on learning outcomes, and language-focused analyses report similar gains, especially in writing. Effects are consistently stronger when use is structured by a teacher.
What did this review find about feedback?
Adaptive feedback was the standout application: instant, level-appropriate comments on writing and speaking that no teacher with 150 students can match by hand. Automatic essay scoring and generated writing prompts showed early promise in the studies surveyed, and both have matured since.
Why was a famous ChatGPT study retracted?
Wang and Fan’s 2025 meta-analysis of 51 studies was retracted in April 2026 because discrepancies in the analysis undermined the editors’ confidence in its conclusions. It is a reminder to cite the careful pooled evidence, not the loudest number.
What is the biggest risk of AI in the language classroom?
Overreliance: students outsourcing the productive struggle that builds skill, plus model bias and confident errors slipping past uncritical readers. The review’s ethics discussion and the 2026 evidence agree that teacher mediation is the control for both.
One question for the road: if your school handed every language student an AI tutor tomorrow, what is the first rule you would set for using it? Mine is simple: show me the draft you started with. Tell me yours in the comments.
References
Bereiter, C., & Scardamalia, M. (2014). The Psychology of Written Composition. Routledge.
Brierley, J., & Kempe, A. (2021). Artificial intelligence, education and the future of work: A transdisciplinary approach to research and practice.
Brown, H. D. (2007). Principles of language learning and teaching (5th ed.). Pearson Education.
Brown, R. (2021). Implications of GPT models in Second Language Acquisition. Linguistics Today, 35(2), 78-95.
Chapelle, C. A. (2001). Computer applications in second language acquisition: Foundations for teaching, testing, and research. Cambridge University Press.
Crossley, S. A., & McNamara, D. S. (2017). Computers and Second Language Learning: Past, Present and Future. John Benjamins Publishing.
Dai, W., Yang, Y., & Xue, G. (2022). How can pre-training models advance natural language processing? A survey. Journal of Computer Science and Technology, 36(2), 202-225. doi: 10.1007/s11390-020-00832-9
Duolingo. (n.d.). Company. https://www.duolingo.com/company
Ellis, R. (1994). The Study of Second Language Acquisition. Oxford University Press.
Gass, S. M., & Mackey, A. (2007). Input, interaction, and output: An overview. A guide to research on second language teaching and learning, 3-16.
Gredler, M. E. (2016). Learning and instruction: Theory into practice (7th ed.). Pearson.
Goo, J., & Paek, J. (2021). Understanding the relationship between artificial intelligence and language learning: A review of the literature. Educational Research Review, 31, 100348. doi: 10.1016/j.edurev.2020.100348
Hsu, Y. C., Lin, Y. H., & Huang, Y. M. (2021). The effects of personalized AI-assisted English vocabulary learning on EFL learners’ vocabulary acquisition and retention. Journal of Educational Technology Development and Exchange, 14(1), 31-54.
Johnson, L. (2018). The efficacy of AI and GPT models in Second Language Acquisition: A review of the literature. Applied Linguistics Review, 9(1), 27-45.
Jones, M. (2019). Ethical considerations in the use of AI and GPT models in Second Language Acquisition. TESOL Journal, 20(3), 45-62.
Krashen, S. (1982). Principles and practice in second language acquisition. Pergamon.
Kuo, I. H., & Chuang, T. Y. (2021). Impact of artificial intelligence on language learning: An updated review. Education and Information Technologies, 26(1), 1105-1126. doi:
Lai, C., Zhu, W., Chen, C., & Li, W. (2017). Using AI technology to enhance teaching and learning: A case study on improving English grammar proficiency. Journal of Educational Technology Development and Exchange, 10(1), 1-14.
Larsen-Freeman, D., & Long, M. H. (1991). An introduction to second language acquisition research. Longman Publishing Group.
Long, M. (1996). The role of the linguistic environment in second language acquisition. In W. Ritchie & T. Bhatia (Eds.), Handbook of second language acquisition (pp. 413-468). Academic Press.
Liu, D. (2020). The impact of artificial intelligence on second language teaching and learning. Journal of Education and Training Studies, 8(8), 129-139. doi: 10.11114/jets. v8i8.4927
Lyster, R. (2007). Learning and teaching languages through content: A counterbalanced approach. John Benjamins Publishing.
Meskill, C. (2018). Language learning with technology: Methods and approaches. TESOL Press.
Rosetta Stone. (n.d.). Our History.
Shen, L., Wang, T., Su, J., & Zhu, X. (2019). Generating Writing Prompts for ESL Students with Generative Pre-trained Transformer. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (pp. 5628-5637).
Smith, J. (2020). How AI and GPT technology can enhance language learning outcomes. Journal of Applied Linguistics, 45(2), 123-140.
Smith, P. Q. (2020). The impact of AI and GPT models on second language acquisition. Journal of Second Language Acquisition, 3(2), 32-47.
Sun, C., & Chen, Y. (2021). The impact of artificial intelligence on language learning. Open Journal of Social Sciences, 9(1), 40-45. doi: 10.4236/jss.2021.91004.
Sweller, J. (1988). Cognitive load during problem solving: Effects on learning. Cognitive Science, 12(2), 257-285.
Usesignhouse. (2022, December 30). Duolingo Stats & Demographics for 2022 [Infographic]. https://www.usesignhouse.com/blog/duolingo-stats
VanLehn, K. (2011). The relative effectiveness of human tutoring, intelligent tutoring systems, and other tutoring systems. Educational Psychologist, 46(4), 197-221.
Vesselinov, R., & Grego, J. (2019). Duolingo effectiveness study. Duolingo, Inc.
Warschauer, M., & Healey, D. (1998). Computers and language learning: an overview. Language Teaching, 31(02), 57-71.
Zhang, M., Zhang, J., & Liu, H. (2020). Feedback-based GPT-2 in the writing domain: An exploration of student writing performance and feedback quality. Journal of Educational Technology Development and Exchange, 13(1), 41-54.
Zhang, Y., Li, H., Li, X., & Li, S. (2020). Natural language processing and language learning: A survey. IEEE Access, 8, 41423-41438.
As an Amazon Associate, we earn from qualifying purchases.
Keep the watchdog independent
If this article saved you money, time, or a headache, if you enjoyed it, or if you want to support local investigative journalism, you can buy me a coffee by card. No account needed, and Ko-fi takes no cut, so it all goes to the work.
Prefer steady support? Join the Patreon for monthly backing.
