The research question
Artificial intelligence can return writing feedback almost immediately, but speed alone does not establish that students learn more. A recent classroom study asked whether combining AI-generated suggestions with comments from fellow students could improve both engagement and writing performance.
Published in Frontiers in Psychology on 21 September 2026, the research involved 122 Chinese students learning English as a foreign language. The investigators compared an integrated AI-and-peer feedback approach with conventional peer feedback.
The findings favoured the combined approach across several measures of student engagement and writing quality. Yet the study also illustrates why educational technology should be evaluated as part of a teaching process, rather than as a replacement for teachers or classmates.
The distinction is important for universities adopting generative AI. An application can propose revisions, but students still need to interpret feedback, judge its relevance and learn to make independent writing decisions.
How the study was designed
The researchers used a mixed-methods design that combined performance assessments and questionnaires with interviews and reflective journals. The intervention lasted ten weeks within a twelve-week study period.
Participants were divided into two groups of 61. The experimental group received feedback integrating AI and peers, while the comparison group relied on conventional peer feedback.
Both groups completed writing assessments before and after the intervention. The study also included five writing assignments, allowing researchers to observe work over the teaching period rather than relying on a single final task.
Engagement was measured across behavioural, affective and cognitive dimensions. These categories reflect different ways of participating in learning, from completing tasks to feeling invested in the process and thinking carefully about revisions.
To add qualitative detail, sixteen students from each group participated in semi-structured interviews and maintained reflective journals. Those accounts helped the researchers interpret how students experienced the feedback process.
The design offers more information than a survey of opinions alone. It examines writing outcomes alongside the ways students interacted with feedback, although it remains situated in one particular educational context.
What the researchers found
Students receiving integrated AI-and-peer feedback demonstrated significantly higher behavioural, affective and cognitive engagement than those using conventional peer feedback, according to the study’s reported analyses.
The integrated group also showed greater improvements across four assessed writing dimensions: task achievement, coherence and cohesion, grammatical range and accuracy, and lexical resource.
The gains were not identical across these dimensions. That matters because stronger vocabulary use, more accurate grammar and better organisation are related but distinct writing skills.
The published abstract reports statistically significant advantages but does not provide a single universal percentage improvement for the intervention. It would therefore be misleading to describe the method as increasing writing ability by a fixed amount.
The sample was evenly divided, with 61 participants per group, giving the investigators a direct comparison between two instructional approaches during the intervention.
These results suggest that the value of AI feedback may depend on how it is combined with human discussion and revision. The study did not compare every possible AI-only, teacher-only or peer-only arrangement.
Why engagement is more than participation
Behavioural engagement concerns observable actions such as taking part in feedback activities, completing writing assignments and revising drafts. Students can appear busy, however, without deeply processing advice.
Cognitive engagement refers to the mental effort devoted to evaluating suggestions and understanding why a change may improve a text. This is especially important when an AI system provides fluent but potentially imperfect recommendations.
Affective engagement captures feelings such as interest, confidence or frustration during the learning process. Students who find feedback approachable may be more willing to revise, while students who feel overwhelmed may disengage.
The three dimensions should not be collapsed into a single claim that students simply liked AI. The researchers investigated whether the integrated feedback approach related to several distinct forms of learning engagement.
Peer discussion may also make feedback more socially meaningful. When classmates explain or challenge suggestions, students have an opportunity to articulate their reasoning instead of accepting revisions automatically.
At the same time, greater engagement does not guarantee long-term mastery. Follow-up assessments would be needed to establish whether advantages persist once the intervention ends.
Why AI and peers may complement each other
AI tools can supply rapid responses and multiple suggestions, potentially helping students notice problems that might otherwise remain unaddressed between lessons.
Peers can contribute another perspective by explaining how a reader understands the argument, asking questions and negotiating alternative word choices. Their comments may also reflect the immediate expectations of a shared course.
Combining these sources creates an opportunity for comparison. Students can consider whether automated advice agrees with peer feedback, identify disagreements and decide which revisions serve the writing task.
This comparison may encourage active judgement, a skill that becomes more important as writing tools become increasingly capable of producing polished text.
Nevertheless, the proposed mechanisms should be distinguished from measured results. The study reported advantages for the integrated approach, but not every possible explanation was experimentally isolated.
AI suggestions can contain inaccuracies or alter intended meaning. Human peers can also provide incomplete or mistaken advice. A robust classroom process needs to teach students how to assess both.
What the findings mean for educators
For lecturers, the study provides evidence that a structured combination of AI and peer review deserves consideration in second-language writing courses.
Implementation should start with learning objectives. If students are expected to improve argument structure, for example, feedback activities should require them to explain how revisions change the logic of their work.
Educators can ask students to document which suggestions they accepted, rejected or modified. Such reflection may help distinguish genuine learning from simply transferring automated text into a final submission.
Peer review remains important because writing is a form of communication. Receiving feedback from actual readers can reveal confusion that a language model’s fluent response may not expose.
Assessment practices also need to recognise the difference between the quality of a submitted document and the student’s independent ability to produce and revise writing.
Schools and universities should consider access to technology, privacy, academic integrity and the reliability of feedback systems before adopting a particular tool.
The research does not demonstrate that institutions should require a specific commercial AI platform. It evaluates a teaching configuration, not a universal product recommendation.
Important limitations
The sample consisted of 122 Chinese learners of English as a foreign language. Their language needs, institutional context and familiarity with educational technology may differ from those of students elsewhere.
The intervention covered ten weeks, with observations collected within a twelve-week study period. The results cannot establish whether the same advantages remain years later.
The comparison was between integrated AI-and-peer feedback and conventional peer feedback. Without additional groups, it is difficult to determine precisely how much of the advantage came from AI, the integration process or other features of the learning activities.
Self-reported engagement measures can be influenced by expectations and students’ perceptions of a novel teaching method. Interviews and journals provide useful detail but are also shaped by participant interpretation.
Writing assessments offer a more direct outcome than attitudes alone, yet their interpretation depends on the scoring process and the kinds of writing tasks selected.
Results from one course setting should not automatically be generalised to younger children, native-language writers or students studying subjects outside language education.
Future studies could compare AI-only, peer-only and integrated feedback under similar conditions, track performance after the intervention and test whether students transfer improvements to independent writing.
The wider significance
The findings contribute to a growing discussion about how generative AI should be used in education. The central question is not merely whether AI can generate a useful correction, but whether its use helps students develop stronger skills.
In this study, 61 learners received integrated AI-and-peer feedback and 61 received conventional peer feedback. The integrated approach was associated with stronger engagement and improvements across four writing dimensions.
That is promising evidence for carefully designed collaboration between students and AI systems, not proof that automation can replace the social and instructional elements of learning.
The most useful lesson may be that feedback works best when students remain active decision-makers. Technology can expand the number of suggestions available, while peers and educators can help turn those suggestions into understanding.
Source Information
Study: AI-peer integrated feedback in second language writing classes: exploring students’ engagement and writing performance
Author: Wen Liu
Journal: Frontiers in Psychology, volume 17
Published: 21 September 2026
Methods: Mixed-methods classroom intervention; 122 learners; two groups of 61; ten-week intervention; pre- and post-tests, five writing assignments, engagement questionnaires, interviews and journals.
