The opinion polling industry is currently facing a transformative shift as artificial intelligence begins to replace traditional, labor-intensive research methods. With response rates to standard surveys plummeting from 30% in the 1990s to less than 5% today, firms are turning to AI agents to gather high-quality insights. Startups like Naratis in France and Outset in the US are leading this charge, promising qualitative research that is significantly faster, cheaper, and nearly as accurate as human-conducted interviews.
Naratis, founded in 2025 by engineer Pierre Fontaine, uses a sophisticated system of AI agents to conduct political and commercial polling. When an AI agent calls a participant, it doesn`t just ask them to check boxes; it engages in a natural, business-like conversation. During this process, three separate background agents monitor the dialogue: one ensures the participant stays on topic, another probes for deeper emotional responses, and a third verifies that the respondent is a real human and not a bot. Fontaine claims this method is ten times faster and 90% as accurate as traditional human polling.
The accuracy of AI-driven polling has been backed by recent academic findings. According to a study published in Royal Society Open Science in March 2026, AI-generated estimates for the 2020 US Presidential election were found to be on par with high-quality traditional models like those from FiveThirtyEight. The study demonstrated that Large Language Models (LLMs) can extract structured survey data from unstructured digital traces on social media, producing timely area-level estimates that correct for the biases found in online samples.
While the commercial sector has embraced AI-moderated interviews for user research and brand testing, the use of AI in politically sensitive polling remains controversial. Established firms like Ipsos are utilizing AI for behavioral analysis—such as analyzing videos of consumers at home—but they remain hesitant to use "synthetic people" or AI-generated respondents in official political surveys. The industry is currently experimenting with "digital twins," which are virtual models of real individuals designed to simulate how they might react to specific campaign slogans or policy changes.
Despite the potential for increased efficiency, the rise of AI polling brings ethical concerns regarding the manipulation of public opinion. Critics argue that while AI can predict outcomes based on existing data, it may struggle with the unpredictable nature of human emotion during high-stakes political events. However, as the cost of traditional polling continues to rise and participation falls, AI appears to be the only viable path forward for understanding the complex web of global public opinion in real-time.
