Pre-election polls shape how voters understand electoral competition, yet news outlets vary widely in how they present them. To assess how design and content choices in poll reporting affect audience perceptions, we fielded a large visual conjoint experiment during the 2025 German federal election (N = 23,300). Participants were presented with one of 160 versions of the same poll, randomly varying reporting formats and technical elements such as reference values, uncertainty information, and sampling details. Across all comparisons, any visualisation — bar charts, dot plots, or line graphs — were rated as substantially more attractive and more trustworthy than text-only formats, which still dominate real-world poll reporting, especially among less statistically literate participants. The effects of technical features were more mixed: some improved trust, others reduced perceived clarity, and few altered expectations about the state of party competition. These results show that how polls are communicated meaningfully shapes their reception and provide evidence-based guidelines for designing clearer, more trustworthy reporting of public opinion signals.