A Machine Relations analysis of 75,000 brands has produced the clearest citation signal ranking to date. YouTube mentions correlate with AI citation share at 0.737. Branded web mentions sit at 0.664. Branded anchor text at 0.527. Brand search volume at 0.334-0.392. Traditional backlinks come in last at 0.218 — the signal the industry has optimized for is the weakest predictor of AI visibility.
The Muck Rack data behind this adds weight: 84% of all AI citations trace to earned media sources across 25 million analyzed links. Only 38% of Google AI Overview citations come from pages in Google's own organic top 10. The implication is that the highest-leverage AI citation work happens off the brand's own site — in editorial coverage, Reddit threads, YouTube, and industry roundups — not in technical on-page optimization.
Brand stature tiers make the gap concrete: global household names appear in 73% of relevant AI answers, established mid-market brands in 44%, and niche brands in 11%. Pages with 19 or more distinct data points earn 2-3x more citations than equivalent pages with fewer. Best-of ranked listicles are the most-cited content format at 21% of AI citations — ahead of how-to guides and opinion pieces.
A new arXiv survey covering 2023-2026 GEO research identifies two factors as consistently robust across studies: topical relevance and context position. Position within the retrieval context window proved more influential than content rewrites. Specific techniques — adding statistics (+25.9%), quotations (+27.8%), and source citations (+24.9%) — show moderate to strong support, but only after a document is already retrieved.
The nuance: these gains are conditional. An end-to-end pipeline test found that body-only content optimization reduced top-20 presence by 9% and final citations by 6%, because changes that improve a page's apparent answer quality can hurt the upstream retrieval signal. The survey also flags a citation fidelity problem — only 51.5% of AI-cited sentences were fully supported by the cited source.