This note examines a structural gap in the EU's approach to governing algorithmic recommendation systems. While the AI Act and the Digital Services Act together constitute one of the most detailed digital governance frameworks in existence, their classification logic relies primarily on sectoral criteria and content-based risk thresholds. This leaves a descriptive blind spot about systems whose governance relevance derives not from what they produce in any given context, but from the structural position they occupy across interconnected digital ecosystems. The note develops the concept of relational reach as an analytical complement to existing regulatory vocabulary. Relational reach names how algorithmic effects travel across systems, through shared technical architecture, cross-platform human learning, and social-graph-mediated amplification. The third mechanism (collaborative filtering as invisible social amplification) is analytically central, because recommendation systems enlist users' social graphs as amplification infrastructure without the knowledge of users, researchers, or regulators. That asymmetry between what the platform sees and what governance can observe is itself the governance problem. The argument does not call for new instruments or expanded mandates. It proposes a descriptive toolkit with four interrelated dimensions of observation, that allows regulators and researchers to articulate the position of high-reach AI systems with greater precision within the interpretative space of existing EU frameworks. This note forms part of the CARA Initiative (Cohesion and Amplification Risk Assessment), an ongoing empirical research programme.
Nicko Nogués (2026) studied this question.