Artificial intelligence cannot replace real humans in opinion polls because its answers differ significantly from actual respondents, with gaps up to 40 percentage points.

Analysis

The idea that AI could instantly generate nationwide poll results is attractive because it promises lower costs and faster turnaround, especially as traditional response rates decline. However, the Pew study shows that AI models rely on outdated training data and embed stereotypical assumptions, leading to large systematic errors on timely and nuanced issues. Readers may overlook that AI lacks the lived experience and contextual awareness of real people, which is why the model mis‑read recent events and over‑generalised demographic attitudes. Recognising these limitations matters because policymakers, journalists, and marketers might otherwise base decisions on misleading synthetic data, potentially distorting public discourse and resource allocation. As AI technology improves, future models may narrow the gap, but for now the evidence warns against treating AI‑generated poll answers as a substitute for genuine human surveys.