Artificial intelligence is fundamentally reshaping how pollsters collect public opinion, with a French start-up called Naratis spearheading efforts into what promises to be a quicker, more cost-effective alternative to traditional survey methods. The company, established in 2025 by 28-year-old engineer Pierre Fontaine, deploys conversational AI agents to conduct in-depth interviews with respondents, replacing the time-consuming work that has long characterised qualitative research. Rather than requiring respondents to select options, Naratis’s AI engages citizens in genuine dialogue intended to examine not just what they think, but how they think. The technology claims to deliver results ten times faster and at a fraction of the expense of traditional survey methods, whilst maintaining 90 per cent accuracy—a major advancement as the polling industry grapples with declining participation levels and growing public distrust.
The Rise of Dialogue-Based Polling
At the heart of Naratis’s innovation lies a deceptively simple concept: substituting the transactional nature of traditional surveys with authentic dialogue. When a participant answers the phone, they encounter a young, brisk AI voice posing open questions about politics, society and their personal views. Rather than simply recording answers, the system conducts real dialogue. Three separate AI agents work simultaneously behind the scenes—one ensuring the respondent stays on topic, another probing for further understanding when answers seem superficial, and a third confirming the person is authentic and not a bot exploiting the system. This multi-layered strategy transforms polling from a routine box-ticking task into something considerably sophisticated and insightful.
The productivity gains are remarkable. In the past, qualitative research necessitated lengthy and demanding work: gathering small cohorts of respondents, performing individual interviews, converting discussions into text, and then analysing responses for trends and insights. Naratis compresses the timeframe using what Fontaine describes as “parallelisation”—numerous AI tools running interviews in parallel rather than people conducting work one after another. A study that once took weeks and many thousands of euros can now be accomplished in one or two days. Feedback frequently returns by the next day, allowing campaigns, government bodies and groups to react to unfolding events and evolving public sentiment almost in real time, substantially altering the tempo of public opinion analysis.
- AI agents carry out concurrent interviews across several respondents
- Instant analysis identifies superficial answers needing deeper exploration
- Fraud detection stops automated responses and inauthentic answers from distorting data
- Results delivered within hours instead of multiple weeks of standard research
Pace and Effectiveness Reshape Survey Methodology
The survey sector confronts an existential crisis. Response rates have plummeted from more than 30% in the 1990s to below 5% today, according to AI consultant Stéphane Le Brun. This dramatic decline has created a vicious cycle: lower participation mean higher costs per finished questionnaire, which in turn renders studies less representative of the wider public. Confidence in polling has eroded in turn, with many regarding polls as unreliable or intrusive. Against this backdrop, AI-powered conversational polling provides a potential solution, potentially reversing years of falling participation by making the research process itself more engaging and interactive.
Naratis claims its AI-driven approach achieves results that are “10 times faster, 10 times more cost-effective and 90% as accurate as human polling.” These numbers, if independently verified, would represent a seismic shift in the way organisations grasp public opinion. The cost savings alone are game-changing: a thorough qualitative investigation that once required tens of thousands of euros and several weeks of work can now be completed for a fraction of the cost within days. This democratisation of access could allow smaller organisations, local campaigns and community groups to undertake thorough opinion research formerly available only to well-funded institutions.
Parallelisation: The Key Breakthrough
The technical advance enabling these gains is elegantly straightforward: parallel processing. Rather than human interviewers performing interviews sequentially—one conversation after another—AI agents operate in parallel across many respondents. This increase in throughput without equivalent expense growth reshapes the economics of polling. Where conventional research methods required substantial commitment, AI-driven approaches reduce timeframes whilst lowering expenses, enabling companies to obtain rich, detailed understanding on demand.
Precision Assertions and Industry Scepticism
Naratis’s contention that its AI methodology achieves 90% accuracy matching human polling has understandably prompted examination from experienced analysts. The polling industry, founded on decades of procedural improvement, remains wary of claims that machine learning can mirror the refined assessment of skilled researchers. Critics question whether conversational AI can truly detect the delicate interpersonal signals, hesitations and body language that experienced practitioners use to investigate further respondent motivations. The company has failed to produce independent research substantiating its accuracy claims, leaving independent verification pending.
Beyond concerns about accuracy, industry observers worry about possible prejudices embedded within AI systems themselves. If the algorithms underlying Naratis’s conversational agents are developed using biased data sets or programmed with untested presumptions, those flaws could systematically distort results across thousands of interviews. Additionally, respondents may change their conduct when interacting with machines rather than humans, either becoming more candid or more cautious depending on their comfort with technology. These psychological and technical variables are largely unexamined territory, and their impact on polling reliability stays unclear.
- Independent verification of precision assertions is still pending from established research institutions
- Possible systematic prejudices could systematically distort results across extensive artificial intelligence survey programmes
- Human-AI interaction effects may alter how respondents articulate authentic views and beliefs
The Synthetic Data Dilemma
As AI polling grows, a troubling question surfaces: how will the public and regulators tell apart genuine human responses and synthetic data produced by the very systems conducting the polls? The speed and efficiency that makes AI polling attractive also generates possibilities for tampering. If an dishonest actor were to bolster actual responses with artificially generated ones, the final dataset could appear statistically robust whilst showing little similarity to actual public opinion. The system’s lack of transparency worsens the concern—most voters would find it difficult to grasp how algorithms synthesise and validate responses, making it challenging for them to trust the findings influencing political debate.
Naratis maintains its systems include fraud prevention systems, with one AI agent tasked with determining if respondents are genuine humans or bots. However, this safeguard itself is contingent on AI making determinations about AI, producing a self-referential flaw. As conversational systems become increasingly sophisticated, differentiating real human exchanges from synthetic responses may become technically impossible. The polling industry has historically possessed public trust partly because its approaches are conceptually simple—people respond to surveys, results are tallied. AI polling risks compromising that clarity, substituting intelligible methods with opaque algorithms that most cannot effectively scrutinise.
Confidence and Compliance Concerns
Regulators in Europe are only now come to terms with AI’s role in political polling and opinion research. Currently, minimal safeguards govern how AI systems gather, analyse and present polling data. Without robust regulatory frameworks, the industry faces a loss of public trust if synthetic data infiltrates published results or if computational biases consistently compromise findings. France’s data protection authorities and the European Union’s AI Act implementation bodies must without delay establish standards securing openness, auditability and oversight in AI-enabled polling work before the technology becomes embedded in political processes.
The Combined Landscape of Market Research
Despite the gains in efficiency AI polling provides, industry experts suggest that human and machine-driven studies will likely coexist rather than one displacing the other entirely. Traditional polling methods have endured decades of examination and remain embedded in political institutions, regulatory frameworks and public understanding. Companies such as Naratis acknowledge that AI performs exceptionally well in speed and cost efficiency, yet human interviewers bring invaluable subtlety—the ability to read subtle emotional cues, adapt questions intuitively and establish connection that encourages candid responses. A measured strategy combining both methodologies could produce deeper understanding whilst maintaining the openness voters increasingly expect from research influencing electoral discourse.
The transition to hybrid models, however, requires thoughtful balance. Pollsters must establish clear protocols for the circumstances under which AI data should be given weight alongside conventional methods, and the manner in which results should be communicated to guarantee public comprehension of which methods yielded which conclusions. Training a new generation of researchers to operate proficiently alongside AI systems presents another challenge, as does developing ethical guidelines that oversee the technology’s application. If managed thoughtfully, this development could revitalise opinion research by increasing speed and accessibility whilst safeguarding the human expertise and moral stewardship that protect democratic discourse.