Researchers have demonstrated that large language models can be used to completely undermine online survey data by acting as synthetic respondents who evade standard detection tools with near-perfect success. The AI tool mimics human behavior by adopting demographic personas, generating realistic answers to open-ended questions, simulating reading and response times, and even typing keystroke-by-keystroke with minor typos and authentic-looking errors. In testing, this synthetic respondent bypassed attention check questions, behavioral flags, logic puzzles, and other safeguards, achieving a 99.8% detection avoidance rate in trials. The study warns that this vulnerability threatens the reliability of survey-based research because a small number of fake responses could sway poll results or distort scientific findings, and that current recruitment and validation methods may no longer be sufficient to ensure data integrity.
