Campaigns have traditionally relied on polling, message testing, media behavior, field conversations and election results to understand what voters are hearing and responding to. As voters increasingly encounter political information through individualized AI interactions, some signals may become harder to observe even as campaigns gain new tools for modeling voters and anticipating reactions. What happens when campaigns become better at simulating voters at the same time they become less able to directly observe them?