Task-differentiated Generative AI tool selection among pre-service teachers: An observational study

Abstract

Generative AI tools now populate higher education at a scale that has shifted the central challenge from simple adoption to strategic, task-aligned selection. Pre-service teacher education in the Philippine context remains underrepresented in longitudinal, task-level research on this phenomenon. This descriptive-quantitative study tracked AI tool selection by 103 pre-service teachers enrolled in action research courses at a Philippine state university across 25 sequential course tasks distributed over three academic phases within one semester. A structured, course-embedded observational tracker captured tool selections, use intensity ratings, multi-tool strategies, and AI-assisted content estimates. ChatGPT dominated every phase with 1,341 total uses, yet selection was meaningfully differentiated by task type. Elicit and Consensus AI-led literature synthesis tasks; Canva AI and Gamma AI dominated presentation work; Grammarly AI, Quillbot, and Writefull clustered around manuscript editing. A chi-square test confirmed a statistically significant association between academic phase and tool selection (chi-square [18] = 360.40, p < 0.001, Cramer's V = 0.20). AI use intensity increased monotonically across phases (Early: M = 2.00; Late: M = 3.86), and multi-tool usage rose from 43.1% to 58.6%. These patterns indicate that pre-service teachers exercise purposive, task-calibrated AI selection consistent with emergent domain-specific AI literacy. The findings carry direct implications for teacher education curricula and institutional AI policy within the Philippine higher education context.