This study compares two designs of automated guidance for short essays in an online thermodynamics unit. Students are prompted by automated guidance to either "revisit" evidence in a dynamic model prior or to plan "writing" changes prior to revision. In this paper we specifically examine how receiving either type of guidance influences science learning, as well as the critique and revisions that students give to another student's essay response. In a school with a large population of English language learners, those assigned to the "writing" condition showed greater pre to posttest knowledge gains than those in the "revisiting" condition. Students in the "writing" condition also gave critiques that included more concrete scientific ideas than those in the "revisiting" condition.