French Startup Naratis Pioneers AI Conversations for Political Opinion Polling in 2025
A French startup, Naratis, is revolutionising qualitative opinion polling by employing AI agents to conduct in-depth conversations with respondents. Founded in 2025 by engineer Pierre Fontaine, Naratis claims its method is “10 times faster, 10 times cheaper and 90% as accurate as human polling”.
Instead of traditional, labour-intensive one-on-one interviews or focus groups, Naratis uses conversational AI to explore how individuals form opinions, rather than simply recording their responses. This allows for studies that once took weeks and tens of thousands of EUR# to be completed within days, enabling clients to react to political events in near real-time.
Polling Industry Faces Declining Trust
The introduction of AI polling coincides with a challenging period for the industry. Response rates to surveys have plummeted from over 30% in the 1990s to below 5% today, according to AI consultant Stéphane Le Brun. This decline makes traditional polling more expensive and less representative, contributing to public scepticism regarding poll accuracy.
While firms like Ipsos integrate AI into market research, using it to analyse video footage of behaviour or create “digital twins” and “synthetic people” for hard-to-reach groups, caution prevails in political polling. OpinionWay CEO Bruno Jeanbart stated his firm would “never publish an opinion poll based on AI-generated data” due to trust concerns.
Potential Benefits and Risks of AI Integration
Advocates argue that AI can reduce certain biases, as respondents may be more candid with a machine, potentially addressing issues such as the consistent underestimation of far-right support in French polls. However, significant risks persist. AI systems can “hallucinate,” generating plausible but incorrect answers, or produce “common sense” responses that undermine genuine public sentiment.
The use of synthetic data, where responses are generated rather than collected, raises fundamental questions about what is truly being measured and how such data should be interpreted. Industry observers expect that regulatory bodies, particularly in Europe, may eventually prohibit the publication of polls based on entirely synthetic data. For now, a hybrid approach, combining AI speed with human oversight, is viewed as the most viable path forward for an industry under increasing economic and public scrutiny.
