Measuring Human Behavior at the Speed of AI and Digital Platforms

Artificial intelligence and digital platforms increasingly shape human behavior online, yet researchers’ capacity to measure that behavior has not kept pace with the speed or scale of these changes. In the post-API era, platform data have become costly and closely guarded assets; trace data are often inaccessible or revocable; conventional data collection cycles can be too slow for rapidly emerging phenomena; and studies that do obtain platform data are generally confined to a single platform and what they are willing to share – offering limited insight into an increasingly fragmented, cross-platform information environment. This dissertation develops and applies a rapidly deployable, responsive, granular, and relatively inexpensive measurement infrastructure based on online surveys for studying human behavior across digital platforms and AI-mediated environments. Across four chapters, I demonstrate how large-scale opt-in online surveys can facilitate rapid data collection, allow researchers to design questions around their own research aims rather than the constraints of platform-provided data, and be paired with methods ranging from econometric inequality measurement and longitudinal modeling to privacy-preserving LLM-assisted response analysis and randomized experiments. The first chapter, published in the Journal of Survey Statistics and Methodology, establishes the methodological foundation by evaluating whether an opt-in online survey can generate useful population estimates for specific outcomes and comparing its performance with official administrative data and a high-quality probability survey. The second chapter begins demonstrating the broader applications of this infrastructure by measuring inequalities in two forms of online participation –content consumption and production– using Gini coefficients, Lorenz curves, Theil decomposition, and top-share statistics. It establishes a standardized benchmark for comparing participation across 12 major platforms in the United States. The third chapter investigates whether platform-specific information trust can help explain why some users produce more content while others remain silent or leave platforms entirely, using cross-platform regression models and longitudinal analyses that test whether lower trust predicts subsequent platform exit. It also employs a privacy-preserving, LLM-assisted pipeline to analyze one of the largest collections of self-reported platform-exit explanations, identifying the reasons and platform-specific patterns associated with quitting behavior. The fourth chapter extends the same data collection infrastructure to experimental research, proposing a randomized experiment to examine whether and how the timing of AI disclosure affects persuasion. Together, the dissertation argues for and demonstrates the value of deploying large-scale online surveys and combining them with formal inequality measurement, longitudinal models, privacy-preserving AI-assisted response analysis, and randomized experiments to produce timely and comparable evidence about human behavior in digital environments increasingly transformed by AI.