Can AI stop AI from faking surveys?
Research Paper Presented at ESOMAR Congress 2025
In collaboration with SurveyTester, five distinct bot types were engineered — ranging from simple clickers to advanced AI agents with memory and personas. Together, we stress-tested leading in-survey quality checks to explore one critical question: Can AI-powered defenses really stop AI-driven fraud?
Introduction
Research agencies are losing an estimated £209 million per year in the UK alone due to poor data quality. This includes costs from re-fielding, lost staff time, license fees, and compensating clients for unusable results (Harding, 2025). It is rare to hear someone publicly report that flawed market research led to poor business decisions. Yet Tia Maurer, Group Scientist at Procter & Gamble, did exactly that. Speaking at the ReDem Quality Day last year, she revealed how fraudulent survey data resulted in the launch of an oral care product that ultimately failed, resulting in significant financial losses and reputational damage for the company (Maurer, 2024). This example underscores a critical vulnerability in today’s market research industry: The growing threat of survey fraud.
Some may argue that survey fraud is nothing new. But why, then, is it suddenly such a hot topic? Why are we seeing a surge in conference talks, articles, and new fraud detection tools? Why has a global industry body, the Global Data Quality Initiative, been formed to tackle this issue? And why, ultimately, was this paper accepted for publication?
The answer is simple: Survey fraud has changed. Radically.
That’s why this paper begins by examining the shifts that have reshaped the fraud landscape. It then explores panel providers, their clients and AI as the three forces that help explain why, despite growing awareness, the problem remains largely unsolved. Finally, it presents the first experimental evaluation of whether AI-powered solutions are capable of stopping AI-generated survey fraud.
This study was conducted jointly by ReDem and SurveyTester to evaluate whether AI-powered tools can reliably detect AI-based survey fraud. While ReDem develops and deploys fraud detection systems, SurveyTester created bots designed to simulate realistic respondent behavior. By testing these bots against an AI detection pipeline, we aimed to create a controlled, adversarial environment, allowing us to examine a critical question for the future of the industry: Can AI detect AI? Or in other words, can AI be the solution to a problem AI itself has helped create?
