Rapid developments in educational technology awaken hopes for making learning more engaging and effective. At the same time, Cognitive Load Theory stresses limitations of human cognitive architecture and urges developers to design learning tools that help learners optimize their mental capacities. In a 1.5-month long study we investigated tertiary students’ use of an AI-enriched digital biology book comprising a 5000-concept knowledge base and algorithms that offer the possibility to ask questions and receive answers. Our aim was to identify and investigate differences between three types of cognitive load (CL), namely, intrinsic (ICL), germane (GCL) and extraneous (ECL), as well as their correlation with learning gain and usability perception. Findings show that non-optimal design, which draws learners’ cognitive resources from the task is linked with a lower learning gain and user satisfaction. The study contributes to new approaches on differentiating between cognitive load types and their relationship with learning from digital tools. The findings also emphasize the importance of optimally designing emerging educational technologies.