Junior high school student perspectives on the use of ChatGPT in music education

Synthesis: Weng and Chiang asked 508 Taiwanese junior high school students what shaped their attitude toward using ChatGPT for lyric learning, and answered by welding experiential marketing (XM) — Sense, Feel, Think, Act and Relate — onto the Technology Acceptance Model. Structural equation modeling shows the integrated model explaining 53% of the variance in attitude toward use and 63.4% in perceived enjoyment, with XM → PEOU the strongest path (Std. β = 0.630, p < 0.001). Enjoyment mediated the link from experience to attitude (PE → ATU, Std. β = 0.369, p < 0.001), and intention to use followed attitude (b = 0.716, p < 0.001). The paper argues that rational acceptance constructs under-explain emotion-driven coursework, so emotional involvement and prompt design are instructional variables rather than backdrop. It is cross-sectional and single-region, and never names the ChatGPT model version students used, so the emotional mechanism is well-evidenced while causal and technical claims stay modest.

Key Findings

1. The integrated XM–TAM model explains 53% of the variance in students' attitudes toward adopting ChatGPT for lyric learning, with XM → PEOU the strongest path (Std. β = 0.630, p < 0.001).
2. XM and PEOU jointly explain 63.4% of the variance in perceived enjoyment, and PE predicts attitude toward use (Std. β = 0.369, p < 0.001).
3. PU, jointly explained by XM and PEOU with R² = 0.384, significantly affects ATU (b = 0.318, p < 0.001); ATU then predicts intention to use (b = 0.716, p < 0.001).
4. The intervention was four procedures over 150 min — prompt demonstration, generation and comparison, rewriting and personalization, sharing and feedback — mapping onto Feel, Think, Sense/Act and Relate.
5. Data came from 508 valid responses to 550 questionnaires (92.36%); female students were 59.06% of the sample and most respondents were 15 years old (322, 63.39%).
6. Most respondents used the free version of ChatGPT (414, 81.5%) while 94 (18.50%) used the paid version; the highest item mean was 4.42 and the lowest 3.58.

Acceptance as an affective process

The paper's move is to argue that the Technology Acceptance Model is a rational account of adoption, emphasizing perceived usefulness and ease of use, while music coursework is grounded in emotional resonance. Experiential marketing supplies the complement: Sense, Feel, Think, Act and Relate operationalize "learning experience" as something measurable, grounded in Kolb's experiential learning cycle. The argument rests on mediation, not a direct effect. The study reports the direct XM → ATU path as non-significant (p = 0.191), concluding that experiential appeal works through enjoyment and usefulness. One inconsistency deserves note: the paper also reports ATU as significantly affected by XM (b = 0.125, p = 0.046), so mediation is the headline but the coefficients do not fully agree with it.

Designing the lyric-learning session

For replication, the intervention was four procedures over 150 min. Step 1 modeled structured prompts for lyric generation, targeting emotional engagement (Feel). Step 2 had students produce three distinct lyric versions under specified instructions, prompting comparative analysis (Think). Step 3 had students revise lyrics to incorporate personal experience (Sense and Act). Step 4 had groups exchange brief feedback on emotional resonance and linguistic fluency (Relate). All constructs were assessed immediately after the session, keeping responses tied to authentic instruction and limiting memory-related bias. The single-session design means reported attitudes reflect one encounter with the tool, not a sustained course.

How the measurement holds up, and what it does not settle

Confirmatory factor analysis under maximum likelihood estimation, followed by a second-order CFA, produced composite reliability from 0.796 to 0.903 and average variance extracted from 0.557 to 0.691, with loadings of 0.616 to 0.943. Fit after the Bollen–Stine correction was strong: normed χ2 of 1.094, RMSEA of 0.014, CFI and TLI of 0.997. The instrument behaves well, which is why this page carries medium confidence; the ceiling comes from the design and from under-specified technology. No ChatGPT model version is named, 81.5% of students used the free version, the collection window is only "a three-month period", and the outcome is intention rather than behavior.

What this means for practice

• Instructors. Design the emotional arc of the lesson before the technical demonstration: the strongest path (XM → PEOU, β = 0.630) links experiential appeal to ease of use.
• Instructors. Run the session as staged work rather than open access, since the paper credits each stage with a distinct experiential dimension.
• Instructors. Treat enjoyment as a learning signal, not a distraction: PE carried the strongest mediation to attitude.
• Curriculum designers. Budget for access equity, because 81.5% of this sample was on the free tier.
• Researchers. Record the model and version, session dates, and access tier, none recoverable here.

Limitations

• The design is cross-sectional, which the authors say captures covariate relationships but does not allow precise causal inference among constructs.
• The data are single-source self-reports, and the outcome is behavioral intention rather than actual usage, which the authors note may be constrained by contextual or policy factors.
• The sample came mainly from junior high school students in a single country or region, which the authors say limits generalizability to other cultural contexts or educational systems.
• The ChatGPT model version is never specified and the collection window is only "a three-month period"; with 81.5% of students on the free version, reported enjoyment depends on undated tool capability.

Citation

Weng, S.-S., & Chiang, H.-C. (2026). Junior high school student perspectives on the use of ChatGPT in music education. Computers and Education Open, 11, 100387.