Nonlinear Interaction with LLMs: Enhancing Divergent and Convergent Processes in Complex Information Tasks

Large language models are increasingly used to support complex information tasks. However, most systems still rely on linear conversational interfaces that may not align well with workflows requiring parallel exploration and iterative integration. To examine whether nonlinear interaction improves task performance, we developed a graphical interface that supports both divergent and convergent processes throughout the task. In a laboratory experiment (N = 21), participants completed creative writing tasks and multi-document synthesis tasks using both the nonlinear interface and a linear dialogue platform. Results showed that nonlinear interaction improved creativity and writing quality in creative writing tasks, and informativeness and conciseness in synthesis tasks. Behavioral and subjective measures also indicated greater cognitive engagement in the nonlinear condition. However, the nonlinear interface received lower usability ratings, suggesting a trade-off between cognitive benefits and interaction overhead. Overall, these findings suggest that nonlinear workflows better support complex information tasks, despite added usability costs.