The potential use, influence and impact of health research is seldom fully realised. This stubborn problem has caused burgeoning global interest in research aiming to address the implementation ‘gap’ and factors inhibiting the uptake of scientific evidence. Scholars and practitioners have questioned the nature of evidence used and required for healthcare, highlighting the complex ways in which knowledge is formed, shared and modified in practice and policy. This has led to rapid expansion, expertise and innovation in the field of knowledge mobilisation and funding for experimentation into the effectiveness of different knowledge mobilisation models. One approach gaining prominence involves stakeholders (e.g. researchers, practitioners, service users, policy-makers, managers and carers) in the co-production, and application, of knowledge for practice, policy and research (frequently termed integrated knowledge translation in Canada). Its popularity stems largely from its potential to address dilemmas inherent in the implementation of knowledge generated using more reductionist methods. However, despite increasing recognition, demands for co-produced research to illustrate its worth are becoming pressing while the means to do so remain challenging. This is due not only to the diversity of approaches to co-production and their application, but also to the ways through which different stakeholders conceptualise, measure, reward and use research. While research co-production can lead to demonstrable benefits such as policy or practice change, it may also have more diffuse and subtle impact on relationships, knowledge sharing, and in engendering culture shifts and research capacity-building. These relatively intangible outcomes are harder to measure and require new emphases and tools. This opinion paper uses six Canadian and United Kingdom case studies to explore the principles and practice of co-production and illustrate how it can influence interactions between research, policy and practice, and benefit diverse stakeholders. In doing so, we identify a continuum of co-production processes. We propose and illustrate the use of a new ‘social model of impact’ and framework to capture multi-layered and potentially transformative impacts of co-produced research. We make recommendations for future directions in research co-production and impact measurement.