Algorithmic Demand Induction in Platform Interfaces: A Computational Audit Framework for Preference-Shaping Recommender Systems

Recommender systems are commonly framed as tools for predicting user preferences and helping users discover relevant products, content, services, or information. Yet contemporary platform environments increasingly combine recommender systems with commercial ranking, engagement optimisation, persuasive interface cues, behavioural feedback loops, and limited user-control pathways. In such environments, recommendations may do more than match existing preferences: they may stimulate, amplify, or redirect user interest toward products, content, services, or ideas that users did not actively seek. This paper introduces Algorithmic Demand Induction (ADI) as a socio-technical concept for examining this shift from preference prediction toward preference shaping. To strengthen the concept as a technical and data-oriented contribution, the paper proposes ADI-TRACE, a computationally implementable audit framework for identifying observable demand-inducing conditions in platform interfaces. ADI-TRACE integrates six dimensions: relevance drift, commercial mediation, persuasive cue load, repetition persistence, control friction, and explanation opacity. These dimensions are operationalised through an ADI Index that can be applied to manually collected or semi-structured platform audit data. The paper does not claim to infer platform intent or prove causal preference change. Instead, it provides a reproducible audit logic for observing and comparing interface-level conditions through which recommender systems may shift from preference matching toward preference shaping.