This preprint presents a short research proposal and statement of interest for the Stanford Summer Institute in Computational Social Science (SICSS). The project examines when Generative Engine Optimization (GEO) tactics in AI-mediated search may become manipulative by increasing perceived credibility without increasing actual evidential support. Generative search systems such as Google AI Overviews, ChatGPT Search, Perplexity, and Bing Copilot increasingly shape how users access information and make decisions. In this new search environment, webpages may be selected, summarized, or cited inside AI-generated answer cards. While some GEO practices may improve clarity, readability, and source traceability, this proposal argues that GEO can become problematic when trust-signaling features, such as citations, statistics, quotations, expert framing, update dates, and confident answer-first formatting, make weakly supported information appear more authoritative than it actually is. The proposed study develops a structured audit framework for examining generative-search answers and their cited webpages. The audit will compare trust-signaling intensity with evidential support quality across platforms, domains, and query types. It focuses on high-trust and decision-relevant domains such as health, finance, shopping, and software/tool recommendation. By analyzing both generated answer cards and cited source pages, the project investigates how credibility cues travel from web content into AI-mediated search outputs and when visibility optimization may become trust miscalibration or an epistemic dark pattern. This proposal contributes to emerging research on GEO, dark patterns, AI-mediated information access, and computational social science methods for auditing digital platforms. It also outlines how SICSS-Stanford could help strengthen the project through feedback on audit design, annotation protocols, coding reliability, and computational measures of trust-signaling and evidential support.
Xinqi Zhang (Sun,) studied this question.