When Large Language Models Become the News Resource: An Analysis From Media System Dependency Theory

In this theoretical article, I examine how news dependency changes when large language model answer engines generate responses from journalism before users open source articles. Journalism produces a public article through editorial synthesis, yet an answer engine can use that article to produce an artificial intelligence (AI) response through which the user understands the news. In Media System Dependency theory, processing precedes audience contact and individual dependency centers on dissemination. During a human–AI interaction, processing can compose the AI response that meets the user's news need, thereby expanding individual resource scope. Repeated interactions create a separate transformation through personalization, the broader process by which user information can shape later outputs. Persistent conversational memory provides one mechanism within that process by carrying information from an earlier interaction into subsequent responses. When chat history materially helps meet the user's news need in a later interaction, the service can become harder to replace, increasing dependency intensity when equivalent context is costly to reconstruct elsewhere. The two transformations remain independent and are empirically distinguishable by identifying whether the original news article or the AI response meets the user's news need and testing whether chat history changed the AI response. Users can still share journalistic sources and overlapping issues even when AI responses diverge within relations organized by the same provider infrastructure. Users can therefore rely on the same provider without receiving the same response.