This research explores how AI-powered “digital representatives” might address Rotterdam’s democratic deficit, where citizen input loses richness as it moves through municipal processes. In Rotterdam, this disconnect manifests in participation disparities (voter turnout ranging from 60% in affluent areas to 21% elsewhere) and declining institutional trust. Through iterative development and evaluation, the study created AI personas capable of maintaining citizen perspectives in policy discussions. These digital representatives integrated speech recognition, Large Language Models, and voice synthesis to participate in deliberations about municipal issues. Key findings demonstrate that while LLMs can consistently represent citizen viewpoints, their authenticity depends entirely on input quality. Representatives derived from rich interview transcripts preserved nuance and influenced policy discussions, while those created from typical municipal documentation proved inadequate. Voice interaction generated stronger engagement than text interfaces, and individual exploration enhanced subsequent group discussions. Evaluation with Rotterdam civil servants revealed both promising capabilities and significant concerns. Digital representatives successfully challenged assumptions about community needs, but participants expressed legitimate concerns about accuracy and the risk of replacing direct citizen engagement with technological simulation. The research’s primary contribution is not the technology itself but revealing critical deficiencies in how municipalities capture and process citizen knowledge. Current documentation practices often fail to retain the experiential context that gives citizen input its value. Addressing Rotterdam’s democratic challenges requires transforming how institutions value citizen perspectives—preserving context rather than pursuing abstraction. Digital representatives cannot solve Rotterdam’s democratic challenges alone but offer a means to sustain citizen perspectives throughout policy processes, especially for voices easily lost in current systems. Their most potent function may be provoking necessary institutional conversations about how citizen knowledge is valued—helping fill, not just simulate presence in, the empty chair.