Steering operations, such as menu navigation and drawing, constitute a fundamental GUI interaction. The conventional Steering Law estimates movement times by accumulating the local width of the path, but it suffers from limited predictive accuracy for paths with varying curvature. In contrast, humans tend to adjust their speed by anticipating the path ahead (e.g., when driving). This study proposes a new Preview-based Model, which integrates geometric properties with visual constraints further along the path. Our experimental results show that our model achieved a higher adjusted R^2 (0.983) on curve-included paths than existing models. Furthermore, a simplified formulation without parameter tuning also exhibited comparable accuracy (adjusted R^2 = 0.972), confirming its effectiveness as a practical difficulty metric.