Ranking signals & measurement
The signals that decide whether you hold a citation or lose it, and how to actually measure any of this when your analytics can't see most AI traffic.
The weighted formula, the sample size that makes it stable, and benchmark bands by vertical. The metric that replaces the screenshot.
Mentions, citations and recommendations are three different AI-search signals with three different fixes. The taxonomy, the math, and what the data shows.
AI engines never read your page, they read chunks of it, scored alone. How chunking, embeddings, vector drift and reranking decide what gets cited, and how to write for it.
How human rater preferences in RLHF and DPO quietly decide which brands AI recommends, and the GEO playbook to win Share of Model.
AI sessions convert at 4.4x organic; GA4 misses about 30% of AI referrers. The prompt portfolio, GA4 dual-setup, and Looker Studio dashboard that prove GEO ROI.
96% of AI Overview citations go to E-E-A-T-trusted sources. How RLHF wired quality into LLMs, and the 90-day framework to close the gap.
Pages are 3.2x more likely to lose AI citations after 90 days. The technical reason why, platform-by-platform heuristics, and the refresh system to fix it.