How ChatGPT Cites Sources: A 2025 Data Study (www.tryzenith.ai)

A September 13–14, 2025 empirical study compared how ChatGPT (with web search enabled) and Google Search cite sources for 80 high-intent, non‑branded B2B queries across three companies (Mindtickle, MotherDuck, Prodigal). Researchers normalized cited URLs to root domains and categorized them. Key quantitative shifts: ChatGPT cites competitor sites 11.1 percentage points more than Google and a company’s own site +3.0 pts; encyclopedic sources jump +3.1 pts and academic +1.4 pts, while industry publications, forums and user‑generated content decline (forums −7.9 pts, industry pubs −4.8 pts). Industry-specific examples show Prodigal’s competitor citations rising from 13% (Google) to 51% (ChatGPT), developer tools retaining more forum presence, and horizontal SaaS favoring vendor voices over individual blogs. Significance: the study signals a structural change in AI authority—models prefer primary (vendor) sources and neutral factual references to avoid intermediary “content‑farm” noise, using forums mainly as supplementary authenticity checks. For AI/ML practitioners and B2B marketers this means rethinking SEO for generative engines: publish authoritative comparisons and encyclopedic, factual explainers, and maintain high‑signal community engagement so models can both verify facts and surface real‑world context. The findings also touch on verification and legal clarity: clearer attribution to primary sources reduces ambiguity but requires continuous measurement as models and citation behaviors evolve.