The labor-intensive process of collecting public opinion is undergoing a transformation as artificial intelligence takes over tasks once performed by human interviewers. A new wave of AI-powered polling firms, including French startup Naratis, claims to conduct qualitative research ten times faster and cheaper than traditional methods, while maintaining 90% accuracy.
Naratis uses conversational AI agents to conduct in-depth interviews. When a participant responds, three separate AI models analyze the answer: one checks relevance, another probes for deeper insights, and a third screens for fraud. This parallel processing allows the company to complete studies in days rather than weeks.
"We don't ask people to tick boxes — they have a conversation with an AI," explains Pierre Fontaine, founder of Naratis. "That means we can explore not just what people think, but how they think — how they build their opinions, and even when those opinions change."
The rise of AI polling comes amid declining response rates for traditional surveys, which have fallen from over 30% in the 1990s to below 5% today. This has made polling more expensive and less representative, fueling public distrust. AI offers a potential solution by enabling faster, cheaper, and more flexible data collection.
Established firms like Ipsos are also integrating AI, using it to analyze video footage of consumer behavior and to create "digital twins" — virtual models of real individuals that can simulate responses. However, in politically sensitive contexts, caution prevails. Ipsos and other firms avoid using AI-generated respondents for political polling, citing concerns about trust and accuracy.
The benefits of AI-driven polling are clear: it reduces certain biases, as people may be more candid with a machine on sensitive topics. In France, this has helped uncover support for the far-right that traditional polls underestimated. However, risks remain. AI systems can "hallucinate" plausible but incorrect answers, and synthetic data raises questions about what is actually being measured.
"The goal is end-to-end automation, but today it would be unsafe and socially unacceptable to remove humans entirely," says AI consultant Stéphane Le Brun. Human oversight remains essential for validating results.
The future of polling likely lies in a hybrid approach, where AI augments human efforts rather than replacing them. For political polling, the boundary between collecting real responses and generating simulated ones will be crucial. As the industry pushes toward greater automation, how the technology is used, explained, and regulated will determine whether it restores or further erodes public trust.