Aligning ChatGPT with E-Portfolio Assessment as EFL Learning Model: Its Effect on Students' Speaking Performance and Feedback Literacy

Synthesis: This quasi-experimental study (N=60, first-semester vocational EFL students at Politeknik Negeri Bali) is the first to empirically test the combined effect of aligning ChatGPT with e-portfolio assessment (the "CEA" model) on both speaking performance and feedback literacy, which prior research had studied only separately. One-way MANOVA found a large simultaneous effect (Wilks' λ = 0.476, p < .001, partial η² = 0.524): the experimental group outperformed the control on speaking performance (M = 82.733 vs 78.833) and feedback literacy (M = 79.666 vs 71.762). Thematic analysis of 27 interviews surfaced seven themes, including digital peer synergy, digital confidence Scaffolding, tech-enhanced autonomy, and metacognitive awareness enhancement. The authors argue that e-portfolios support speaking ability and the metacognitive ability to use feedback, and that pedagogically structured AI integration adds value beyond raw AI access.

Key Findings

• Large simultaneous effect: MANOVA on the two dependent variables was significant — F(2,57) = 31.313, p < .001 (Wilks' λ = 0.476), a large overall effect (partial η² = 0.524). The CEA model significantly improved students' speaking performance and Feedback Literacy together.
• Speaking performance: univariate test F(1) = 48.554, p < .001, partial η² = 0.456 (large). Post-test means: experimental M = 82.733 (SD 2.381) vs control M = 78.833 (SD 1.931).
• Feedback literacy: univariate test F(1) = 16.135, p < .001, partial η² = 0.218. Post-test means: experimental M = 79.666 (SD 9.830) vs control M = 71.762 (SD 4.418).
• Seven qualitative themes from 27 interviews: digital feedback orchestration, systemic feedback information construction, development of adaptive learning strategies, digital peer synergy, digital confidence scaffolding, tech-enhanced autonomy, and metacognitive awareness enhancement.
• Feedback as durable record: students reframed Feedback as consultable data (stored on Padlet) rather than a temporary correction, supporting the "processes feedback information" and "commits to feedback as improvement" dimensions of Feedback Literacy.
• Structured integration matters more than access: consistent with prior work (Teng, 2024; Teng & Huang, 2026), the authors emphasize that how AI is incorporated into Pedagogies and Teaching Strategies outweighs the mere availability of the technology — CEA is an AI-grounded value-add to traditional e-portfolios, not AI access alone.

Study Design & Method

• Design: explanatory sequential mixed-methods design combining a quasi-experiment (pretest–posttest control group, one factor with two dependent variables) with qualitative semi-structured interviews.
• Participants: 60 first-semester students (2 of 6 classes, ~18–20 yrs, all non-native speakers ~9 yrs of English, CEFR B1) in the International Business Management program, English for Interpersonal Communication course, Politeknik Negeri Bali. Cluster random sampling; 27 experimental students purposively interviewed via WhatsApp.
• Intervention (8 weeks): Experimental group used the CEA model — Padlet as the primary e-portfolio environment (video submissions, peer comments, collaborative Feedback), with ChatGPT as a secondary individualized feedback system (voice-mode practice, idea drafting, pronunciation, archiving transcripts and self-reflections via Google Form). Control group used a conventional e-portfolio on Google Drive with lecturer-only feedback and no structured generative AI.
• Instruments: speaking tests scored on fluency, accuracy, and complexity (rubrics adapted from Housen & Kuiken 2009; Ogawa 2022); a 14-item feedback literacy self-rated questionnaire adapted from Molloy et al. (2020). Inter-rater reliability strong (Cohen's κ 0.761–0.840).
• Analysis: one-way MANOVA with Wilks' λ after checking normality/homogeneity/covariance assumptions; qualitative data analyzed via Braun & Clarke's (2019) six-phase thematic analysis with dual coding, peer debriefing, and audit trail.

What this means for practice

• Instructors. Pair ChatGPT with an e-portfolio instead of offering raw AI access: the CEA model's simultaneous gains in speaking performance and Feedback Literacy came from pedagogically structured integration, not availability.
• Instructors. Design Feedback as a durable, consultable record so students orchestrate and reconcile input from the lecturer, peers, and AI — the behavior this study tied to feedback literacy.
• Instructors. Use voice-mode practice and archived self-reflection to create a low-anxiety environment that raises willingness to communicate rather than heightening performance anxiety.
• Assessment designers. Make Feedback Literacy an explicit intended outcome and pair e-portfolio evidence with AI feedback to support Authentic Assessment and progress monitoring.
• Designers. Purpose-build the AI architecture of the assessment task; unstructured access carries dependence and "high-tech plagiarism" risks, and how technology is incorporated matters more than whether it is available.

Limitations

• The experimental and control groups used different e-portfolio systems (Padlet vs. Google Drive), so Padlet's additional functionality and social features may independently explain the engagement gains — a confounding variable the quasi-experimental design cannot fully account for.
• The sample was 60 first-semester students at one vocational institution (Politeknik Negeri Bali), which the authors state limits generalizability to these participants.
• The control condition was not AI-free, since ChatGPT was widely available to students outside class, so it represents typical e-portfolio practice rather than a pure baseline.
• The qualitative themes rest on 27 purposively interviewed experimental students only, and no follow-up beyond the 8-week intervention tested whether the increases were maintained.

Citation

Laksana, I.P.Y., Ratminingsih, N.M., Santosa, M.H., & Kusuma, I.P.I. (2026). Aligning ChatGPT with e-portfolio assessment as EFL learning model: its effect on students' speaking performance and feedback literacy. Technology in Language Teaching & Learning, 8, 103616. CC BY-NC 4.0.