Audience engagement has become a key concept in contemporary discussions on how news companies relate to the public and create sustainable business models. These discussions are irrevocably tied to practices of monitoring, harvesting and analyzing audience behaviours with metrics, which is increasingly becoming the new currency of the media economy. This article argues this growing tendency to equate engagement to behavioural analytics, and study it primarily through quantifiable data, is limiting. In response, we develop a heuristic theory of audience engagement with news comprising four dimensions—the technical-behavioural, emotional, normative and spatiotemporal—and explicate these in terms of different relations of engagement between human-to-self, human-to-human, human-to-content, human-to-machine, and machine-to-machine. Paradoxically, this model comprises a specific theory of audience engagement while simultaneously making visible that constructing a theory of audience engagement is an impossible task. The article concludes by articulating methodological premises, which future empirical research on audience engagement should consider.