Can AI improve the accuracy of opinion polls?
Opinion polling is undergoing significant changes with the introduction of artificial intelligence (AI). The question arises: can AI improve the accuracy of opinion polls? AI offers a cheaper and faster way to collect people’s opinions, but its impact on the quality and reliability of polling data is still being evaluated.
AI-driven qualitative polling: a new approach
Traditionally, qualitative opinion research involves small groups or one-on-one interviews with respondents recruited through panels. These interviews are time-consuming and costly, often taking weeks to conduct and analyze. A French company called Naratis is pioneering the use of conversational AI to transform this process.
Instead of asking respondents to tick boxes, Naratis uses AI agents to engage people in conversations. This method explores not only what people think but also how they form their opinions and when those opinions change. According to Pierre Fontaine, the founder of Naratis, this AI-driven approach is “10 times faster, 10 times cheaper and 90% as accurate as human polling.” What once took weeks and tens of thousands of euros can now be completed in a day or two, with responses often gathered in under 24 hours.
The speed advantage comes from the ability of AI to conduct many interviews simultaneously, a process called “parallelisation.” This allows clients to react to events almost in real time, a significant improvement over traditional methods.
Challenges and limitations of AI in polling
Despite the benefits, AI polling faces several challenges. Response rates to surveys have dropped sharply over the years, making polling more expensive and less representative. AI may help address this by enabling faster and cheaper data collection, but concerns remain about accuracy and trust.
Critics point to past polling failures, such as the unexpected outcomes of Brexit and the 2016 US presidential election, though these mainly involved quantitative polling. Qualitative research, which focuses on understanding opinions rather than predicting outcomes, may be less affected by these issues.
AI systems can also produce errors known as “hallucinations,” where they generate plausible but incorrect answers. They may default to “common sense” responses based on typical assumptions rather than capturing the true diversity of opinions. This is problematic because polling aims to reflect what people actually think, not what is commonly assumed.
Another concern involves the use of synthetic data and digital twins—virtual models or generated profiles based on real-world patterns. While these tools can help study small or hard-to-reach groups, they raise questions about what is truly being measured and how such data should be interpreted.
Industry perspectives and future outlook
Established polling firms are incorporating AI in various ways. For example, Ipsos uses AI extensively in market research, analyzing video footage of respondents to observe behavior directly rather than relying solely on self-reported data. AI is also applied to social media analysis and experiments with digital twins and synthetic people.
However, in politically sensitive polling, caution remains strong. Firms like Ipsos and OpinionWay avoid using AI-generated respondents for political surveys. OpinionWay’s CEO, Bruno Jeanbart, emphasizes that they would never publish polls based solely on AI-generated data due to trust concerns.
AI-driven polling offers clear advantages: it is faster, cheaper, more flexible, and can reduce certain biases. For instance, people may be more candid with a machine interviewer than with a human, especially on sensitive topics. This could help address issues such as the consistent underestimation of far-right support in French polls.
Nevertheless, trust and regulation are major issues. The introduction of AI, particularly in generating data, could intensify political scrutiny. Some experts expect that countries like France may eventually prohibit the publication of polls based on synthetic data.
Experts agree that human oversight remains essential. Stéphane Le Brun, an AI consultant, notes that while the goal is end-to-end automation, it would currently be unsafe and socially unacceptable to remove humans entirely from the polling process. Human validation and responsibility are crucial for maintaining quality and trust.
For now, the future of opinion polling is likely to be hybrid. AI will expand the scope of polling by enabling large-scale conversational surveys, integrating social media data, and delivering faster insights. Techniques like digital twins and synthetic data may find niche applications, especially in market research, but political polling will likely maintain a clear boundary between augmenting human data and simulating it.
Companies like Naratis are betting on transforming how respondents are heard—turning surveys into conversations and conversations into data at an unprecedented scale. Whether this shift will restore or further erode trust in polling depends largely on how AI is used, explained, and regulated, rather than on the technology itself.