This article introduces the concept of "co-impact" to characterise the complex and dynamic process of social and economic change generated by participatory action research (PAR). It argues that dominant models of research impact tend to see it as a linear process, based on a donor-recipient model, occurring at the end of a project following the take-up and use of findings. PAR challenges this approach, as impact is embedded in cycles of the action research process; the distinction between researchers, research informants and research users is blurred; and micro process-based impacts, including changes in the thinking and practices of co-researchers, are as significant as findings-based changes in policy and practice. A conceptual framework is developed, based on a three-fold distinction between "participatory," "collaborative" and "collective" impact. This is applied to a case study action research project, "Debt on Teesside," working with low-income households in North-east England. The project is analysed in terms of participatory impact (e.g. developing skills of participating households, mentor-researchers, and university staff); collaborative impact (e.g. findings-based changes in thinking, policies and practices of advice, community finance and housing agencies, and local authorities resulting from collaborative research); and "collective impact," adapted from the field of social interventions, which involves organisations collectively targeting specific actions based on research (e.g. changing policy and practices of lenders and government relating to high-cost loans).