Political polling is at a crossroads. Long relied on telephone surveys and mail-in questionnaires, the field now confronts steep declines in response rates and growing public scepticism. Pew Research found that in 1997 roughly one-third of US respondents answered polls; by 2018 that figure had shrunk to just 6 percent, a trend replicated worldwide and seriously undermining traditional polling methods. This crisis has sparked interest in artificial intelligence as a radical alternative. Harvard’s Ash Center points to AI chatbots that instantaneously scan social media, online discussions, and demographic data to summarise public sentiment, offering a potential replacement for slower, more expensive traditional surveys . In one notable study, researchers used a large language model to simulate thousands of demographics-driven human responses. Their findings showed correlations above 85 percent with real-world data on issues like abortion, though the method stumbled over newer policy matters such as emerging positions on foreign conflict. Proponents argue AI could revolutionise polling by running continuous, targeted snapshots of public opinion. Instead of asking a limited sample to react, campaigns might deploy a chatbot to ask and receive instant feedback and calibrate messaging accordingly. In practical terms, AI systems analysing public discourse have already helped predict voter trends days ahead of traditional polls, as seen in historical studies tracking X activity. Yet this shift raises red flags. AI-driven polling depends on opaque systems whose training data and biases remain hidden. While systems are improving at self-flagging uncertain responses, they risk amplifying misinformation and misinterpreting nuance . The danger lies in treating AI outputs as definitive public sentiment rather than invitations to further scrutiny. Broader reflections from the Ash Center highlight a cautious optimism: fear of AI disruption has not fully materialised, but its impact is nuanced. In the 2024 global election cycle—dubbed the largest in history—AI proved influential in campaign operations, translation efforts, and misinformation control, not as an election disruptor. Echoing this point, experts observe that AI amplifies existing political tools, making them more potent rather than introducing entirely new dynamics . Campaign teams are already integrating AI in targeted outreach. The Siena College Research Institute, for instance, deployed an AI chatbot named Engage in 2024 to interact with “persuadable” voters—asking questions and even responding on behalf of respondents. AI bots helped uncover the subtle factors behind voter thinking, demonstrating how campaigns might replace or supplement human canvassers. Meanwhile, smaller-scale political firms use AI to calibrate messages based on real-time data, reflecting growing confidence in machine inference. Despite such advances, AI remains a complement, not a replacement. Its capacity for rapid, scalable polling is compelling, but its limits are clear. Demographic forecasting remains less accurate, and AI lags on topics that emerge after model training dates . Also, synthetic media risks—such as deepfakes and misinformation—continue to worry election officials. Policymakers and academics stress that embracing AI in polling must come with guardrails. Disclosure of AI use, algorithmic transparency, and ethical standards should accompany technical experimentation. Without such measures, reliance on AI may corrode trust in democratic processes. The emergence of AI polling marks a turning point for the democratic process. Harvard’s research paints a cautiously optimistic picture: AI may restore sampling representativeness and respond at scale, but only with appropriate transparency and oversight. As traditional polls falter, AI offers a technological path forward if wielded responsibly—and understood as a tool, not a truth....