Vibe-Coded Websites and SEO: Separating the Selection Effect from the Treatment Effect

The analysis finds that observable site characteristics (schema markup, XML sitemaps, rendering mode, backlinks, and domain authority) differ systematically between AI-built and traditional sites. A descriptive Oaxaca-Blinder decomposition, parameterized from aggregate statistics reported in the cited studies, finds that these characteristic differences are sufficient to account for the entire observed SEO gap under baseline assumptions, with the unexplained remainder being small or zero. This does not prove the absence of a treatment effect; unobserved confounders may offset genuine AI-code-quality penalties, but it establishes that the gap does not require a large adverse treatment effect to explain. The paper's core contribution is the decomposition framework itself; the numerical estimates are illustrative, parameterized bounds, not precise measurements. When AI tools are used with proper SEO guidance and editorial oversight, performance approaches parity with human-built sites. The results suggest that platform-level SEO guardrails (server-side rendering defaults, automatic schema generation, built-in sitemap infrastructure) would close most of the gap regardless of user expertise.