AI Tutors, Worksheets, and Task Design: Matching AI Tools to Learning Structure

Abstract

Artificial intelligence is increasingly used by students as a learning resource in STEM courses. However, classroom observations and survey data suggest that unstructured AI use may correlate with reduced student achievement. Two concerns are especially relevant: AI systems can produce incorrect responses in tasks requiring precise validation, and students may use AI in ways that reduce active engagement with course material. This session presents two instructional approaches designed to address these limitations: structured AI-generated worksheets and in-chat AI tutors. AI worksheets generate controlled sets of practice problems within instructor-defined course constraints. In-chat tutors provide interactive practice through problem generation, Socratic questioning, and feedback-driven guidance. The usefulness of each approach depends on the structure of the learning task. Drawing from iterative development and testing primarily in introductory chemistry, we examine four types of tasks: (1) conceptual and explanatory, (2) structured tasks with low ambiguity, (3) multi-step quantitative tasks, and (4) highly deterministic symbolic or structural tasks. Examples from chemical nomenclature, dimensional analysis, stoichiometry, Lewis structures, laboratory writing, and mathematical validation illustrate where different AI approaches worked reliably—and where they did not. Participants will receive ready-to-use prompts for both AI worksheet systems and in-chat tutors, along with implementation strategies and observed limitations. The session concludes with a practical framework for selecting AI tools based on the structure of the learning task rather than disciplinary content alone.