This article explores the governance by algorithms in information societies. Theoretically, it builds on (co-)evolutionary innovation studies in order to adequately grasp the interplay of technological and societal change and combines these with institutional approaches to incorporate governance by technology or rather software as institutions. Methodologically, it draws from an empirical survey of Internet-based services that rely on automated algorithmic selection, a functional typology derived from it, and an analysis of associated potential social risks. It shows how algorithmic selection has become a growing source of social order, of a shared social reality in information societies. It argues that – similar to the construction of realities by traditional mass media – automated algorithmic selection applications shape daily lives and realities, affect the perception of the world, and influence behavior. However, the co-evolutionary perspective on algorithms as institutions, ideologies, intermediaries, and actors highlights differences that are to be found, first, in the growing personalization of constructed realities and, second, in the constellation of involved actors. Altogether, compared to reality construction by traditional mass media, algorithmic reality construction tends to increase individualization, commercialization, inequalities, and deterritorialization and to decrease transparency, controllability, and predictability.