Inside the CITE score: how we measure AI visibility deterministically
The CITE score rates a domain's AI-answer visibility from 0 to 100 across four weighted dimensions: Citation (35%), whether answers actually cite you; Trust (25%), the signals engines use to pick sources; Identity (20%), whether machines can resolve who you are; and Eminence (20%), third-party corroboration. A structural failure caps the verdict regardless of the total.
Why publish the scoring model at all?
Because a score nobody can inspect is marketing, not measurement. The GEO category has an earned skepticism problem: visibility snapshots that cannot be reproduced, and "AI visibility grades" whose inputs are secret. Our position is simple: nothing unmeasured is presented as proof, and the model itself is public. This article is the long-form version of the methodology page.
What does each dimension measure?
- Citation (35%): real buyer queries run against answer engines, each cell recording cited or not cited, dated. When measurement keys are not configured, this dimension is estimated from on-page signals and explicitly flagged as estimated: an estimate is never dressed as a measurement.
- Identity (20%): structured data (Organization, WebSite), entity clarity, canonical consistency. Engines cannot cite what they cannot resolve.
- Trust (25%): AI-crawler access in robots.txt, server-rendered content, freshness, contact and authorship signals.
- Eminence (20%): presence in the third-party sources engines cross-reference: directories, reviews, comparison articles.
The weights are a starting prior, calibrated against observed citation behavior; branded-mention presence correlates with AI Overview visibility more strongly than backlinks do (Zyppy AI citation study).
| Dimension | Weight | Measured how |
|---|---|---|
| Citation | 35% | Live answer-engine queries (or estimated, flagged) |
| Trust | 25% | Crawler access, SSR, freshness |
| Identity | 20% | Schema, entity clarity |
| Eminence | 20% | Third-party corroboration |
What does deterministic mean here?
Deterministic means the same input produces the same score. Technical access, page citability, identity, and trust are computed from fetched pages by fixed rules, so a month-over-month delta in those layers is real movement, not measurement noise. Only the live answer layer varies, and it is labeled as a dated snapshot.
This split is what makes before/after evidence possible. When a fix cycle moves your citability from 69 to 90, that is not an engine mood swing; the same rules scored both runs. And when the live citation layer moves, the dated matrix shows exactly which queries flipped.
What does the score refuse to claim?
No guarantee of rankings, citations, or inclusion in any specific answer: engines change their sources without notice, and anyone promising otherwise is selling weather control. The score also refuses to fill gaps: unmeasured cells are reported as unmeasured, estimated dimensions are labeled estimated, and vetoes (like zero measured citations) cap the verdict even when the weighted total looks healthy.
How do you read your own number?
Run the free report: you get both scores in seconds, each dimension's evidence, and a top-5 fix list ranked by impact over effort. Then the loop matters more than the snapshot: fix, re-measure, and compare deltas. For the strategic picture, start at What is GEO.