This exploratory mixed-methods study investigates the independent and comparative predictive roles of learner aptitude and engagement in ChatGPT-assisted English writing performance among Bangladeshi undergraduates. Drawing on a purposive sample of 45 students enrolled in an undergraduate English degree program, we administered a simplified 10-item aptitude measure (α = .761) derived from the Modern Language Aptitude Test, a 12-item contextualised engagement scale (α = .905) based on the tri-dimensional engagement model, and a writing task scored with a rubric demonstrating excellent inter-rater reliability (ICC = .988). Regression analysis indicated that engagement was a statistically significant predictor of writing performance (β = .320, p = .032, R2 = .102), whereas the simplified aptitude measure was not (p = .369). Qualitative interviews (n = 10) identified iterative revision, critical feedback processing, and metacognitive reflection as characteristic behaviours of highly engaged learners. These findings should be interpreted as preliminary, context-specific evidence suggesting that, within this particular ChatGPT-assisted writing setting, engagement may be a more salient predictor than this operationalization of aptitude in a low-resource educational context. Generalization is substantially constrained by the small purposive sample, the limited scope of the simplified aptitude measure, reliance on self-reported engagement, and the absence of a control group. Replication using larger, more heterogeneous samples, and fully validated aptitude batteries is strongly encouraged.