The Personalized Web

The Personalized Web traces the development of personalization from Web 2.0 recommendation and engagement systems to generative AI. Building on Shoshana Zuboff’s account of surveillance capitalism, the project examines how an information economy organized around behavioural observation, inference, and prediction has intensified as digital services have become more centralized and computational representations of their users more extensive. The key question is how systems originally designed to anticipate what users might click, watch, or buy have acquired a growing capacity to infer preferences, appearance, relationships, and other dimensions of the people they serve.

Current case studies examine beautification apps and social media applications by reconstructing changes on both sides of the interface. Successive application releases and decompiled code provide evidence of front-end design, data collection, and operations performed on users’ devices. Back-end algorithms are less directly observable, requiring machine-learning papers published by technology companies to be read against technical documentation and evidence disclosed in legal proceedings. Reading these sources in relation makes it possible to distinguish what users encounter, what client software can be shown to do, what companies disclose about their server-side systems, and what otherwise inaccessible evidence subsequently brings to light.

Generative AI extends this history as personalization increasingly incorporates direct disclosure. Behavioural traces remain important, but users now also entrust commercial machine systems with personal histories, relationships, ambitions, anxieties, uncertainties, and other information in sustained conversation, increasingly turning to them for advice, companionship, emotional support, and consequential decisions in their lives. Information that earlier systems sought to infer indirectly may now be volunteered and combined with data accumulated elsewhere across centralized services. The project asks what personalization becomes when computational representations draw not only on observed behaviour, but increasingly on what people choose to disclose in sustained interactions with machines.