This paper argues that for the study and facilitation of collaborative learning, existing theories of grounding such as that of Clark and Shaefer  cannot be applied without adjustments. When comparing collaborative learning and conversation, four dimensions can be identified where grounding at a knowledge level differs from the grounding at an utterance level. Firstly, the indirect access and the existence of a range of manifest meanings, poses the need for a notion of `groundedness'. Secondly, we propose providing evidence in `co-referenced actions' to be an important process as well as an additional marker to assess grounding. Thirdly, instead of simply repairing misunderstandings after they arise, `perspective taking' becomes a more prominent mechanism. Fourthly, effort into grounding is turned from needing to be minimised, into needing to be `optimised'.