In Japan's early childhood education and care settings, the piano is commonly used as the primary instrument. However, instruction in music is typically provided by classroom teachers rather than professional musicians, which can result in difficulties with accompaniment. Shinkai (2012) posits that piano education in early childhood teacher training should not be confined to technical proficiency but should embody the overarching concept of "music for childcare." This viewpoint indicates that accompaniment in early childhood settings encompasses not only technical execution but also expressive and pedagogical dimensions. To support independent practice and practical application, the authors initiated the development of an AI-based simplified accompaniment generation tool. In January 2025, the prototype of the tool's primary function—accompaniment simplification—was completed. The current version has undergone substantial simplification with the objective of prioritizing conciseness in the accompaniment. Subsequent enhancements are expected to yield simplified accompaniments customized to distinct performer levels and utilization objectives. To achieve this objective, it is necessary to establish parameters that define the simplification process. To address this, the present study analyzed qualitative feedback from the prototype's subjective evaluation. The participants were 15 instructors at childcare teacher-training institutions who were selected to represent three distinct perspectives: five music instructors (Group A), five non-music instructors with childcare experience (Group B), and five instructors with no relevant musical or childcare background (Group C). The comments were grouped by profile and analyzed using KH Coder (Higuchi, 2014) to extract frequent words. The semantic tendencies of these words were examined in relation to each group's background to identify their evaluative focus. The study revealed three distinct perspectives. The words selected by Group A, including "beginners" and "difficult," reflected the concept of "playability for novices." The words selected by Group B, including "children," "sing," and "listen," alluded to the notion of "ease of listening and singing for children." The words selected by Group C, including "feel" and "like," signaled subjective preference. Future research will focus on developing parameter sets based on these perspectives to enable the system to evaluate AI-generated accompaniments and generate outputs tailored to specific objectives, such as beginner playability, ease of performance for children, or listener preference. This study was supported by JSPS KAKENHI Grant Number JP24K06016, Japan. The authors express gratitude to Kaito Tsukamoto and Mahiro Sone for their contribution to the development of the AI-based simplified accompaniment generation tool.