This mixed-methods study investigates the efficacy and learner experience of using an AI-generated, multi-dimensional scaffold during the pre-writing stage of continuation writing tasks for Chinese English as a Foreign Language (EFL) learners. The quantitative phase employed a between-subjects design (N = 52 undergraduates), comparing an AI Scaffold Group with a No-Scaffold Control Group. Results indicated the scaffold group produced texts with significantly higher holistic quality, fluency, and lexical alignment, though no significant differences were found in grammatical accuracy. The qualitative phase, which is central to this study's purpose of achieving an in-depth understanding, involved stimulated recall interviews and scaffold use logs with a focal sub-sample. A reflexive thematic analysis of this rich qualitative data revealed that learners experienced the scaffold as a tool that: (1) reduced cognitive load by structuring the complex planning process, (2) deepened their comprehension and connection to the source narrative by activating a detailed situation model, and (3) increased confidence and self-efficacy by providing actionable linguistic and structural guidance. The integrated findings substantiate the role of AI-generated scaffolding as a potential instructional tool, demonstrating that its primary benefits are mediated through cognitive and affective pathways that facilitate a more engaged and meaningful interaction with the writing task. This study contributes to the fields of AI in education and L2 writing by delineating not only the outcomes but, more importantly, the lived experience and underlying learner-cognitive processes of AI-assisted planning, thereby highlighting the essential role of interpretive qualitative inquiry in evaluating educational technology.