Will AI lead to more accurate opinion polls?

Artificial intelligence is automating opinion polling, with companies like France's Naratis using conversational AI to conduct qualitative research they claim is faster, cheaper, and nearly as accurate as human-led studies [1]. Naratis replaces traditional, labor-intensive interviews with AI agents that can conduct multiple conversations simultaneously, a process founder Pierre Fontaine calls 'parallelisation' [1]. The company states its method is 10 times faster, 10 times cheaper, and 90% as accurate as human polling, allowing studies that once took weeks to be completed in days [1]. 'We don't ask people to tick boxes - they have a conversation with an AI,' Fontaine explains, emphasizing the method's focus on understanding how opinions are formed [1]. In the United States, start-ups like Outset, Listen Labs, and Hey Marvin conduct similar AI polling in the commercial sphere, though Naratis claims to be a pioneer in applying it to political opinion research [1]. The shift comes as the polling industry grapples with plummeting response rates, which have fallen from over 30% in the 1990s to below 5% today, making traditional methods more expensive and less representative [1]. Established firms are also integrating AI tools. Ipsos uses AI to analyse video footage of consumer behaviour, while the industry experiments with 'digital twins' and synthetic data to study hard-to-reach groups [1]. However, political pollsters remain cautious. Bruno Jeanbart, CEO of OpinionWay, states, 'we would never publish an opinion poll based on AI-generated data,' citing trust concerns [1]. Critics warn AI systems can 'hallucinate' or produce generic 'common sense' responses, while synthetic data raises questions about what is truly being measured [1]. Despite the push for automation, AI consultant Stéphane Le Brun notes that removing human oversight entirely would be 'unsafe and socially unacceptable' [1].