GEO measurement is a leading-indicator game: the 476-site benchmark sits at a median of 75.0, same-day query runs vary by 10-18%, and about 75% of AI Overview sources change week to week, so single checks lie and trend lines tell the truth.
This hub collects GeoXylia's measurement cluster: the benchmark data, the tracking protocols, and the metrics that predict revenue before the clicks arrive. Measurement is the discipline that separates GEO programs from GEO guesses. In our 476-site audit, the median overall score is 75.0 with a mean of 68.4, meaning the average site has clear room to improve, and the dimensions that drive the score are well understood: the citability score correlates 0.961 with overall, technical foundation 0.944, and content 0.937.
Who is this for? Analysts building tracking stacks, marketers reporting AI visibility to executives, and anyone who has opened a GEO dashboard, seen a number, and wondered whether it meant anything. The cluster answers that question with a method: duplicate runs, weekly cadence, and a two-layer metric set.
What is a good GEO score in 2026?
The reference is GeoXylia's 476-site audit (Aug 2026). The median overall score is 75.0, the mean is 68.4, the 10th percentile sits at 56, and the 90th at 81. Vertical medians span 78 for B2B SaaS (n=54) down to 68 for local businesses (n=35), with tech media, education, finance, healthcare, ecommerce, legal, entertainment, and travel filling the middle band between 73.5 and 76.5. Two practical readings follow. First, most sites are in a 56-81 band, so the difference between a poor score and a strong one is roughly the difference between no program and a working one, not a matter of budget. Second, the dimensions correlate tightly with the overall score, citability at 0.961, technical at 0.944, content at 0.937, so a low dimension score identifies where the overall score is leaking. Use the benchmark as a baseline, not a target: above 81 puts you in the top decile.
Why are my citation numbers different on every run?
Because AI answers are genuinely unstable, and the instability is measurable. Same-day query runs vary by 10-18%, so two runs of the same 40 prompts hours apart will disagree on a meaningful share of results. Week to week the churn is larger: about 75% of AI Overview sources change, and engines replace cited domains at scale during updates, with one major model change swapping out over 40% of previously cited domains. The implications are procedural. Run every prompt set twice on the same day and average the results. Track trends over four or more weeks instead of judging single snapshots. Tag every mention as cited, named-but-uncited, or absent, because recognition leads citation. And never change strategy on one run, whatever it shows. Tools that report a stable weekly number are smoothing over the variance; the raw runs are where the signal lives.
What metrics predict revenue before the traffic arrives?
Two layers. The leading layer: citation rate, share of voice, and branded-search volume. Branded-search lift is the strongest available predictor, correlating 0.334-0.392 with AI citations and forecasting B2B pipeline 4-8 weeks ahead, which makes it the metric to watch when AI referrals are still too small to measure. The lagging layer: AI referral traffic and conversion value, which confirm whether leading indicators converted. One warning governs both: 70.6% of AI visits carry no referrer and appear in analytics as Direct, so referral-based numbers undercount the real channel, sometimes by a wide margin. Supplement analytics with citation monitoring, branded-search tracking, and server logs. The framework guides in this cluster walk through the full protocol, including how to run the weekly prompt set and how to reconcile the numbers across tools.
Where to start
Start with the flagship benchmark, State of GEO 2026: What a 476-Site Audit Reveals, then build your tracking stack with the analytics framework, How to Measure GEO Success, and the metrics guide, The GEO Metrics Framework. Establish your baseline against the benchmark with the free audit at geoxylia.com/audit.
Sources: Similarweb · Ahrefs: AI visibility research · Search Engine Land · SE Ranking blog
Disclosure: AI-assisted, human-edited. Statistics verified against the GeoXylia research base.