We compared two forms of automated guidance to support students' understanding of climate change in an online inquiry science unit. For "specific" guidance, we directly communicated ideas that were missing or misrepresented in student responses. For "knowledge integration" guidance, we provided hints or suggestions to motivate learners to analyze features of their response and seek more information. We guided both student-constructed energy flow diagrams and short essays at total of five times across an approximately week-long curriculum unit. Our results indicate that while "specific" guidance typically produced larger accuracy gains on responses within the curriculum unit, "knowledge integration" guidance produced stronger outcomes on a novel essay at posttest. Closer analysis revealed an association between the time spent revisiting a visualization and posttest scores on this summary essay, only for those students in the "knowledge integration" condition. We discuss how these gains in knowledge integration extend laboratory results related to "desirable difficulties" and show how autonomous inquiry can be fostered through automated guidance.