Advances in automated scoring technologies have the potential to support student learning during inquiry instruction by providing timely and adaptive guidance on individual students' responses. To identify which forms of automated guidance can be beneficial for inquiry learning, we compared reflective guidance to directive guidance for student-generated concept diagrams in web-based inquiry instruction. Eleven intact classes were randomly assigned to either a reflective guidance or a directive guidance condition. After creating a concept diagram showing energy flow in life science during the inquiry instruction, the directive group was told specific ways to improve their diagram, while the reflective group was told to revisit a relevant visualization step to locate useful information. The results from the concept diagrams, as well as individual tests, show that both forms of automated guidance helped students add target energy concepts, but reflective guidance was significantly more effective than directive guidance in improving students' coherent understanding of how energy flows in life science. Analyses of log data revealed that the reflective group was more likely to revisit the visualization step as suggested in the guidance, which significantly enhanced student learning. Detailed analyses suggest that revisiting relevant materials to find useful information challenged students to identify gaps in their understanding and distinguish among multiple ideas. This study shows the value of designing reflective automated guidance for helping students engage in evidence-gathering practices and enhance their understanding of scientific concepts. The findings suggest promising directions for the design of automated adaptive guidance to support complex science learning.