Credibility and Persuasion Signals in US Finfluencer Short-Form Video Content: A Multi-Platform Content Analysis Using Large Language Models

Financial influencers increasingly shape retail investment behavior, and regulatory attention to this content has grown across multiple jurisdictions. But systematic content-analytic evidence on how credibility and persuasion signals actually appear in finfluencer short-form video is still sparse. This thesis addresses that gap through a descriptive content analysis of 1,815 videos from 117 US-based finfluencers across YouTube Shorts, TikTok, and Instagram Reels. Each video was coded by an LLM against ten persuasion frameworks drawn from Source Credibility Theory (Ohanian, 1990; Wellman, 2024), Cialdini's principles of persuasion (Cialdini, 2006, 2016), Speech Act Theory (Austin, 1962; Searle, 1969), and the SafePersuasion taxonomy (Kong et al., 2025). These four frameworks were chosen because each captures a structurally distinct component of how a persuasion act is constructed — speaker establishment, tactical pressure, requested-response framing, and evaluability of the persuasion. The sampling is purposive and engagement-stratified, so claims should be read within those bounds. Two contributions follow from this work. The first is a methodological contribution: this study operationalizes ten persuasion frameworks into a structured LLM coding architecture validated against published indicator sources, with an overall error rate of 4.7% across frameworks. The framework-by-framework prevalence map across the three platforms — reported in full in the appendices — establishes the first systematic at-scale baseline for the prevalence of credibility and persuasion signals in finfluencer short-form video content. The second is an empirical contribution: this study identifies, at scale, several patterns in finfluencer content that prior research has not been able to surface. Among the prevalence-map observations, Instagram showed the highest persuasion density on 7 of 10 frameworks, reciprocity signals appeared roughly three times more often on Instagram than on YouTube, formal credentials appeared in only 1.0% of videos despite broader expertise signaling being present in 47.2%, and classical scarcity tactics were nearly absent across all three platforms. Beyond the prevalence map, the thesis reports four exploratory preliminary findings. COMMAND-level directives operate in a distinct persuasion register from milder directive forms, with substantially lower rates of supporting reasoning (40.7% versus 86–92%) and substantially higher co-occurrence with manipulative techniques (44.1% versus 5–13%). Within a given creator's portfolio, videos containing a COMMAND directive receive roughly 37% more views than that creator's median, an effect that does not generalize to milder directive forms. A small subset of the corpus — 80 videos from 44 creators, 4.4% of the dataset — contains manipulative persuasion techniques without any rational persuasion alongside them, and these videos share a recognizable compositional fingerprint visible across multiple independently coded frameworks. The disclosure forms most commonly named in finfluencer regulatory discourse appear in fewer than 2.1% of videos, while substantive caution behaviors that are observable but not always formally regulated appear at several times those rates. These findings are descriptive and exploratory rather than causal, and they suggest directions for follow-up work in regulatory practice and content moderation. To the best of my knowledge, this is among the first studies to apply all four of these theoretical frameworks simultaneously to a finfluencer corpus at this scale.