AI threatens integrity of online research and polling

Artificial intelligence is increasingly capable of simulating human behavior to complete online questionnaires, posing a significant threat to the integrity of survey-based research. Researchers from the IMT School for Advanced Studies Lucca and the University of Cambridge have warned that this has created a “cat-and-mouse” contest between detection tools and adapting AI systems.

Evidence suggests that fraudulent responses could account for between 4% and 90% of data in certain populations. This contamination is particularly dangerous for social science and political polling, where even a small percentage of false data—as low as 3% to 7%—can invalidate statistical conclusions. The rise of autonomous AI agents allows for industrial-scale manipulation on platforms designed for speed and low cost, such as Amazon Mechanical Turk, Prolific, and Lucid.

To combat this, experts suggest analyzing “paradata,” such as typing cadence and keystroke dynamics, to identify non-human patterns. There is also a call to shift toward vetted, probability-based panels with verified identities to prioritize reliability over scale.