Abstract
Generative artificial intelligence (GenAI) is potentially an effective tool for cheating unproctored internet tests (UIT). However, the practical threat posed by GenAI depends not only on its capabilities, but also whether applicants choose to use it and can do so effectively under real, high-stakes testing conditions. We conducted a series of large-scale surveys of US job applicants who completed selection test batteries in 2026 (n = 141,766) to observe the availability of GenAI assistance, decisions to use that assistance, and the extent to which its use yields higher test scores. Nearly half of applicants reported using GenAI during their job search (49%) but were more likely to use GenAI to help write resumes or to learn about jobs than to help during selection testing. Applicants who used GenAI during testing (9% of all applicants) scored lower on GMA (d = -0.28) and most personality traits (ds between -0.51 and -0.29). Effect sizes were consistent in magnitude and direction across job families, job zones, or AI adoption for work. Our findings suggest that the threat of GenAI on UIT depends not only on the capability of GenAI tools, but also on whether applicants choose to use them and can do so effectively under real testing conditions.