MKEO / Metrics
Three monitored quantities plus audit gates. Each is defined so an independent party can reproduce the number from the open-source reference implementation.
Does the model know the entity exists? Ask engines unaided questions of the form "What is <brand>?" (neutral phrasing, no leading hints). Record per engine whether the answer mentions the entity at all, and whether the description is materially correct.
Unit: mentioned / correct, per engine, per probe date
Interpretation: recognition precedes everything — an engine that cannot name you cannot recommend you
Does the model recommend it when a buyer asks a category question? Ask category-intent questions ("best / how to choose / alternatives for <category>") without naming the brand. Record whether the brand appears among the answers, and at what rank.
Unit: present / rank, per engine, per question, per probe date
Interpretation: the commercial metric — this is where AI visibility turns into demand
Where do the target keywords stand in classic search? For each monitored keyword: top-100 presence per engine and region (e.g., Google US desktop, Bing US, Baidu CN). Engines that return no result for a keyword at all are marked as an anomaly, not as "not found".
Unit: best rank per keyword × engine × region, per probe date
| Gate | Meaning | Target |
|---|---|---|
| Orphan pages | pages with fewer than two internal inbound links | 0 |
| sameAs links | entity links from the site's schema to its platform profiles | ≥ 4 |
| Profiles live | platform profiles that exist and are discoverable | ≥ 4 |
| FAQ coverage | key pages carrying answerable question content | site-specific |