French Startup Naratis Pioneers AI Political Polling, Claims 90% Accuracy, Reduced Costs
A French startup, Naratis, has introduced an artificial intelligence (AI) system for political qualitative polling, marking a significant shift in how public opinion is gathered and analysed. Founded in 2025 by engineer Pierre Fontaine, Naratis employs AI agents to conduct in-depth interviews, a process traditionally labour-intensive and costly.
AI Transforms Qualitative Research
Naratis focuses on qualitative research, a segment of polling that has historically relied on small groups and one-on-one human interviews. Fontaine states that their AI system facilitates conversations rather than simple tick-box surveys, allowing for exploration into how opinions are formed and evolve. The company claims its method is “10 times faster, 10 times cheaper and 90% as accurate as human polling”, enabling studies that once took weeks to be completed in days, often with responses gathered in under 24 hours.
This speed is attributed to “parallelisation”, where multiple AI agents can conduct numerous interviews simultaneously, a stark contrast to human interviewers working sequentially. This development arrives as traditional polling faces declining response rates, now below 5% from over 30% in the 1990s, increasing costs and eroding public trust, according to AI consultant Stéphane Le Brun.
Industry Adoption and Remaining Concerns
Established polling firms, such as Ipsos, are also integrating AI into market research, using it to analyse footage of consumer behaviour directly. The technology is also being explored for social media analysis and the creation of “digital twins”—virtual models of individuals—and synthetic data, which generates new profiles based on real-world patterns. These tools could help in studying small or hard-to-reach demographics.
However, the application of AI in politically sensitive polling remains cautious. Ipsos does not use AI-generated respondents in political surveys, and OpinionWay’s CEO, Bruno Jeanbart, affirms that while AI may conduct interviews, they “would never publish an opinion poll based on AI-generated data” due to trust issues. Critics highlight risks such as AI systems ‘hallucinating’ plausible but incorrect answers, or producing generic “common sense” responses rather than genuine public sentiment.
The creation of synthetic data raises fundamental questions about what is truly being measured. Trust is a substantial hurdle, with Jeanbart predicting that countries like France may eventually prohibit the publication of polls derived from synthetic data. While AI offers clear benefits in speed, cost, and data richness, human oversight is considered essential for validating results and maintaining accountability. The future of polling is likely to be a hybrid model, with AI expanding capabilities while human scrutiny remains paramount, particularly in political contexts where accountability and trust are critical.
