AI-generated simulated survey responses without real respondents — are least reliable for the local-specific questions that matter most in electoral contexts. Using synthetic respondents as a substitute for actual polling in high-stakes electoral applications is a methodological failure, not an effi