Policy 'incoherence' describes a lack of joined-up government that contributes to a confusing mix of policy instruments. It is an inevitable feature of multilevel policy-making, in which many actors compete to set the agenda, and 'inequalities policies', such as gender mainstreaming, which contain multiple and often-contradictory aims. This insight may prompt policy-makers to learn how other governments have responded pragmatically, rather than seeking to design abstract mainstreaming policies with unrealistic levels of coherence. Yet, policy learning is a political process characterized by contestation. Many policy-makers compete to define the policy problem, set the parameters for learning and determine which governments should take the lead. Therefore, we ask: How can we use policy theories to facilitate research-informed policy learning under these circumstances? We describe the framework that we developed for the Horizon 2020 project IMAJINE to encourage policy learning in multilevel policy-making systems. We then illustrate its value in a case study of our work for the Scottish Government's National Advisory Council on Women and Girls (NACWG), which asked us to identify lessons from gender mainstreaming policies in other nations. This framework and case study help explain the limited impact of research on policy learning.