CoChat

Generative AI, visual scaffolds, and metacognition in EFL writing

Open access
Chen, Mei-Rong Alice · 2026 · Language Learning & Technology

The DOI matches its Crossref record; 50 of the 51 references it cites were found in a registry or an index; 16 of the 57 sources its claims cite were checked for support.

Abstract

Fostering metacognitive awareness and reflective thinking remains a challenge in English as a Foreign Language (EFL) writing instruction. These cognitive processes are essential for enabling learners to monitor, evaluate, and engage in meaningful self-reflection throughout the writing cycle. Using Flower and Hayes’s cognitive process model and Jasper’s Experience-Reflection-Action framework, this study proposes a Generative AI-assisted Visually Guided Progressive (GAI-VGP) writing approach. This pedagogical design integrates step-by-step visual scaffolds with interactive AI feedback to foreground learner agency and promote reflective engagement. To evaluate this approach, a semester-long quasi-experimental study was conducted with 58 EFL university students. Data sources included pre- and post-writing tests, metacognitive awareness questionnaires, reflective thinking scales, AI chat logs, and reflection notes. The results showed that the GAI-VGP group significantly outperformed the conventional VGP group in writing performance, planning and evaluation strategies, and reflective awareness. Qualitative analyses indicated increased engagement in understanding and reflectively evaluating writing strategies, although critical reflection remained limited across both groups. Although AI significantly enhanced planning and evaluation, monitoring showed no significant gains, with prompt-log evidence revealing a displacement pattern in which learners outsourced real-time regulation to AI validation.

Overview

CoChat Summary

TL;DR

This quasi-experimental study tested a Generative AI-assisted Visually Guided Progressive (GAI-VGP) writing approach with 58 EFL university students over a semester. It found significant gains in writing performance, planning/evaluation metacognition, and reflective understanding/reflection, but no significant monitoring or critical reflection gains, with prompt logs suggesting monitoring was outsourced to AI validation.

Key findings

  • Writing performance: the GAI-VGP group had a higher adjusted post-test mean (67.94) than the cVGP group (60.44), F = 4.46, p < .05, η² = .08.
  • Metacognitive awareness: overall adjusted post-test mean was 3.97 for GAI-VGP vs. 3.64 for cVGP, F(1, 55) = 10.47, p < .05, η² = .16. Planning was significant (Adj.M = 3.97 vs. 3.59, F = 5.53, η² = .09, p < .05) and evaluation was significant (Adj.M = 3.98 vs. 3.01, F = 20.94, η² = .28, p < .001), but monitoring was not (F = 3.67, p > .05).
  • Prompt logs showed planning prompts were most frequent (35), followed by evaluation (26) and monitoring (14); monitoring prompts often took the form of validation-seeking queries, which the authors interpret as a “monitoring displacement” pattern in which learners outsourced real-time regulation to AI validation.
  • Reflective thinking: understanding was significant (Adj.M = 3.93 vs. 3.31, F = 4.49, p < .05, η² = .08) and reflection was significant (Adj.M = 3.93 vs. 3.16, F = 9.96, p < .01, η² = .15); habitual action and critical reflection were not significant (F = 1.54 and F = 2.20, respectively, p > .05). Reflection-note counts were 14 vs. 5 for understanding, 15 vs. 10 for reflection, and 3 vs. 1 for critical reflection.

Methods

  • Quasi-experimental convergent parallel mixed-methods design with 58 second-year English majors (45 females, 13 males; aged 19–21) in an East Asian EFL context; two intact classes were assigned as GAI-VGP (n = 30) and cVGP (n = 28) over an 18-week semester.
  • Data sources included pre/post argumentative writing tests scored with a TOEFL-based analytic rubric (100 points; Cohen’s κ = .75), an 18-item Metacognitive Writing Strategies Questionnaire (Cronbach’s α = .96 pre, .89 post), a Reflective Thinking Questionnaire, AI chat logs, and reflection notes from 20 volunteers.
  • Quantitative analysis used one-way ANCOVA controlling for pre-test scores, with partial η² and Bonferroni adjustments; qualitative data were analyzed through directed content analysis/thematic coding in NVivo 14, with prompt coding κ = .74, reflection-note coding κ = .66, and qualitative coding κ = .82.

Limitations

  • Stated: the EFL context is often teacher-centered, which may predispose students to accept AI suggestions without critical evaluation; reflection-note participation was voluntary, introducing self-selection bias; and “proxy dilution” occurred when a minority of students dominated AI interactions, potentially benefiting high-agency learners and leading to “surface consultation.”
  • Evident from the text: the sample was small (58) and used intact classes, limiting generalizability; recorded AI prompts were relatively limited and used selectively, so gains cannot be attributed solely to AI use; critical reflection remained limited across both groups, and the monitoring displacement pattern suggests the design may not have scaffolded real-time self-regulation during drafting.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.64152/10125/73695

Cites

50 of 51 references verified0 citation errors · 1 not checkable

Statistics

Figures

Figure 1 Framework for the GAI-VGP Model
Figure 1 Framework for the GAI-VGP ModelCC-BY-NC-ND · From the PDF
Image 2
CC-BY-NC-ND · From the PDF
Figure 3 Example of a Learner's Interaction with the Argumentative Writing Chatbot
Figure 3 Example of a Learner's Interaction with the Argumentative Writing ChatbotCC-BY-NC-ND · From the PDF
Figure 4 The Procedure of the Experiment
Figure 4 The Procedure of the ExperimentCC-BY-NC-ND · From the PDF

Verification

Claims the paper makes

13 Valid · 1 Not valid · 21 Not checked · 88 Own results and interpretations

52 more claims · 2 with their cited sources checked

Create a free account to see all 123 claims and to verify the 4 references still open.

See all claims — free account
  • It may state two separate claims as one. 39
  • Probably not a substantial claim. 34
  • The quote only partly states this claim. 5

Create a free account to see these claims.

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • ConsistentF(1, 55) = 10.47reported p < .05 · recomputed p = .002

    “F(1, 55) = 10.47,p < .05”

  • ConsistentF(1, 55) = 4.49reported p < .05 · recomputed p = .039

    “F(1, 55) = 4.49, p < .05”

  • ConsistentF(1, 55) = 9.96reported p < .01 · recomputed p = .003

    “F(1, 55) = 9.96, p < .01”

3 of 3

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 51 references

Research works printed but not linked: 2

  1. Jasper, M. (2013). Beginning reflective practice (2nd ed.). Nelson Thornes.
  2. Scardamalia, M., & Bereiter, C. (1987). Knowledge telling and knowledge transforming in written composition. In S. Rosenberg (Ed.), Advances in applied psycholinguistics: Reading, writing, and language learning (Vol. 2, pp. 142-175). Cambridge University Press.

Metadata

Authors
Chen, M.-R. A.
Affiliations
Soochow University (TW)
Corresponding author
Chen, Mei-Rong Alice
Venue
Language Learning & Technology
Published
2026-06-29
Volume
30
Issue
1
Pages
1-26
ISSN
1094-3501 (ISSN-L)
OpenAlex
W7166587502
Cited by
1
References
49
Citations by year
2026: 1
FWCI
20.73
Type
journal article
Language
en
Open access
Licence
cc-by-nc-nd
Locations
  • Language learning & technology (publishedVersion, cc-by-nc-nd)
  • ScholarSpace (University of Hawaii at Manoa) (submittedVersion, cc-by-nc-nd)
Funders
  • Ministry of Education, India
  • National Science and Technology Council
Topics
  • Social Sciences › Arts and Humanities › Language and Linguistics › EFL/ESL Teaching and Learning
  • Social Sciences › Education › English Language Learning and Teaching
  • Social Sciences › Psychology › Developmental and Educational Psychology › Innovative Teaching and Learning Methods
Keywords
generative AI, metacognition, EFL writing, reflective thinking, AI-assisted writing, learner agency, displacement effects, AI-generated feedback, critical reflection

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.