AI Representation Theory: A Conceptual Framework

Abstract

Generative AI systems increasingly stand between an information environment and the people who query it. Rather than transmitting a message or returning a ranked list of sources, these systems identify an object, select and rank information about it, compress a fragmented information environment, connect the object to other entities, frame the result and generate evaluations or recommendations. The output is a constructed representation. AI Representation Theory is a conceptual framework that treats this AI-constructed representation as a distinct unit of analysis for communication research and for strategic communication practice. It separates four analytically distinct phenomena - the information environment, the AI representation, AI presence and human effects - and specifies why evidence at one level does not transfer to another. The framework defines four constructs — AI representation, the working version of reality, representational fragmentation and AI presence — together with AI-mediated reputation as the evaluative expression of AI presence for subjects capable of holding reputation. Nine propositions are stated with explicit epistemic status, proposed operationalisation and falsification conditions, separating definitional claims from empirically established findings and from untested hypotheses. The framework integrates rather than replaces established work in framing, agenda setting, algorithmic gatekeeping and human-AI interaction, and situates itself against recent audits demonstrating cross-language, run-to-run and temporal divergence in generative systems. Its principal contribution is a vocabulary and an analytical structure within which visibility, representation, fragmentation and reputation in AI-mediated environments can be distinguished, measured and researched separately.