Cooperative learning in net-based scenarios is often conducted using synchronous communication tools such as text-based chat. Knowledge acquisition by exchanging textual contributions frequently leads to suboptimal results, since the discourse suffers from a lack of coordination and from incoherent contributions. What we call learning protocols is a method to improve coordination and coherence in net-based discourses, and thus to increase learning gains. Learning protocols are implemented as system-controlled cooperation scripts: Participants are required to reference their contributions, to indicate the type of each contribution, and to follow a predefined sequence when submitting a contribution. An experimental study is reported presenting evidence which confirms the positive effect of a learning protocol for mutual explanations, especially with respect to the referencing function. However, it turns out that the usefulness of the learning protocol is highly domain specific.