From Computational Turn to Generative Turn: A Systematic Review of LLMs in Communication and Media Studies

Large Language Models (LLMs) have rapidly moved from novelty tools to routine infrastructures for producing and analyzing communication, intensifying the shift of communication and media studies from a “computational turn” to a “generative turn.” Yet field-wide evidence on how LLMs are selected, deployed, and evaluated remains limited. This systematic review maps LLM use across communication scholarship, distinguishing LLMs used as a method from LLMs studied as a subject/topic, and synthesizing task-specific benefits and limitations. We searched Web of Science and Scopus for English-language journal articles related to LLMs and communication and media studies published after the release of ChatGPT. All collected articles were analyzed by mixed methods. We observed that most subfields of communication and media studies showed similar model-use features, with only a small number of robust associations. Publications were dominated by high-income affiliations suggesting an inequality. Model deployment was strongly task-contingent. Transparency and validation practices remain uneven, with only about three-fifths of GenAI-based studies disclosing concrete prompts and just over half reporting human evaluation of model outputs. Across tasks, reported risks concentrated on validity threats and bias. We conclude by proposing a 3M framework to support more transparent, reliable, and context-sensitive LLM-enabled communication and media studies. Theoretically, our findings clarify how the “generative turn” reshapes research by shifting LLMs from purely analytic instruments to sociotechnical infrastructures that also generate communicative content and participate in communicative processes.