CORAL: Framework for Structuring Individual Cognitive Processes as AI Reasoning Processes

Although mainstream AI development emphasizes efficiency by minimizing human effort, often refining “smart” systems that respond effectively even to vague inputs, this trend paradoxically reduces users to passive operators. Furthermore, at the level of custom instructions, prompt engineering techniques have become increasingly sophisticated; however, their primary objective is to control LLM probabilistic behaviors to efficiently obtain high-quality outputs—a goal distinct from bridging users' own thinking processes with AI reasoning processes. This asymmetry in technological advancement, weighted toward AI efficiency, implies that the current human–AI collaboration faces a structural constraint in which the thinking processes of users are not adequately reflected in AI reasoning processes. This study proposes an approach to structure and verbalize individual thinking processes, implementing them as AI reasoning processes alongside the CORAL framework that realizes this approach. As a proof of concept, we implemented the S-OPE configuration using CORAL and conducted a comparative evaluation study across the four conditions. In complex problem-solving tasks simulating real-world scenarios (35-point rubric), the S-OPE configuration achieved significantly higher scores than a custom instruction that combines multiple existing prompt engineering techniques (median 30.7 vs. 25.0, effect size r = 0.83, p < 0.001). Additionally, the ablation analysis confirmed an integration effect that exceeded the sum of the effects of the individual CORAL framework components. This study demonstrates the feasibility of structuring and implementing individual thinking processes in natural languages.