Query Intent, Not Google Rank: What Best Predicts AI Citation Behavior

Abstract

The rapid growth of AI-powered search has spawned Generative Engine Optimization (GEO) advice built on untested assumptions about how AI platforms select citation sources. Practitioners widely assert that Google ranking determines AI visibility, that Reddit confers citation advantages, and that AI recommendations are too inconsistent to optimize for. We tested these claims across four AI platforms (ChatGPT, Claude, Perplexity, Gemini) using a multi-study design combining large-scale query intent classification (n=19,556 queries, 8 verticals), Google rank cross-referencing (120 API queries plus 100 web UI queries against Google and Bing Top-3), server-side fetch verification, and technical analysis of 479 cited/non-cited pages. Our results challenge all three claims. First, query intent emerged as the strongest aggregate predictor of citation source type, with intent distributions varying significantly by vertical, though page-level technical features outperformed intent for individual citation prediction. Second, Google Top-3 URLs predicted AI citations poorly (ChatGPT: 7.8% API, 6.8% web UI), but domain-level alignment was substantially higher (28.7-49.6%), indicating platforms draw from top-ranked domains while selecting different pages. All platforms aligned 4-7x more with Google than Bing, contradicting claims of Bing-based backends. Reddit, despite occupying 38.3% of Google Top-3 positions, received zero API citations but 8.9-15.6% of web UI citations, revealing a channel-dependent paradox. Third, within-platform brand recommendations showed substantial consistency (ChatGPT Jaccard=0.619), while cross-platform agreement was near-random. We also discovered an architectural divide: ChatGPT and Claude perform live page fetches, while Perplexity and Gemini use pre-built indices, with divergent robots.txt compliance. Effective GEO requires intent-aware, platform-specific optimization rather than one-size-fits-all approaches.

Ähnliche Arbeiten