What Rubric checks
Rubric runs 44 scored checks across three pillars, Known, Findable and Trusted, and weights them for six AI engines: ChatGPT, Perplexity, Google AI Overviews, Gemini, Microsoft Copilot and Claude. The checks measure on-page, structural and technical signals that predict citability. Rubric does not measure your backlinks, your off-page authority, or the citations you have already earned.
- →Rubric scores 44 checks across three pillars (Known, Findable, Trusted) and weights them for six AI engines.
- →The score is an estimate from on-page, structural and technical signals. It does not count citations you have already earned.
- →Rubric does not measure off-page authority such as backlinks. Calibrate the estimate against measured citations from Bing Webmaster Tools.
What does Rubric check?
Rubric checks 44 signals that decide whether an AI engine can quote a page. Every check belongs to one of three pillars, and each pillar answers a question an engine asks before it cites you.
- Known asks whether an engine can resolve who published the page: entity, Organization and Person schema with sameAs, content-type schema, canonicals, internal links, entity density and unique titles and meta.
- Findable asks whether an engine can locate a liftable answer: answer-first openers, question and claim-shaped headings, self-contained sections, tables and lists, image alt text, crawler access, an XML sitemap and a fast, indexable page.
- Trusted asks whether the evidence is strong enough to quote by name: substantive content, statistic density, external source links, a named author with E-E-A-T attribution, statistics that carry a source and freshness.
Rubric also checks for an llms.txt file, but that check is informational and is not one of the 44 scored checks, so it never moves your score. For the full breakdown of each pillar, see the-three-pillars.
Which AI engines does Rubric score for?
Rubric scores for six AI engines, because each one values the same signal differently. The biggest difference is JavaScript: some engines never run it, so anything injected by JavaScript, including schema, is invisible to them.
| Engine | JavaScript rendering |
|---|---|
| ChatGPT | Does not run JavaScript |
| Perplexity | Does not run JavaScript |
| Claude | Does not run JavaScript |
| Google AI Overviews | Renders JavaScript |
| Microsoft Copilot | Renders JavaScript unreliably |
| Gemini | Google engine; weights schema heavily |
This is why "schema readable without JavaScript" is a scored check. If your structured data is injected by JavaScript, ChatGPT, Perplexity and Claude never see it. Schema is weighted more heavily for Gemini, so a site that server-renders its structured data has an advantage. Grok is not scored, because its distinctive signal is heavy weighting of recent posts on X, which is off-page and cannot be measured by an on-page audit. Rubric surfaces Grok as an advisory note instead.
What Rubric does not measure
Rubric is an on-page estimate, and it is honest about what it cannot see.
- Off-page authority. Rubric does not measure backlinks, domain authority or brand mentions elsewhere. Even the Compare view, which shows the citability gap against a competitor, is on-page only.
- Earned citations. The score predicts citability from signals. It does not count the times an engine has actually quoted you.
- A guarantee. Rubric is guidance, not a promise of a citation, a ranking or traffic. No single fix carries a guaranteed outcome.
- Machine files. Sitemaps and .xml, .csv, .json and llms.txt files are not scored as pages, so they do not appear in your page list.
Is the score a count of my citations?
No. Your citability score is an estimate, not a citation count. It is built from on-page, structural and technical signals that predict whether an engine will quote a page.
To sanity-check the estimate against reality, export your Bing Webmaster Tools AI Performance data and compare it per URL with your Rubric scores over time. The estimate should track your measured citations. Where a high-scoring page earns few citations, or a low-scoring page earns many, the gap tells you where to look next. For more on this, see estimate-vs-measured-citations.
Where do the 44 checks come from?
Every check maps to a real, testable signal, not a hunch. Examples include Schema.org types for entity and content markup, the 40 to 70 word answer window an engine can lift, a freshness window of twelve months, and render parity so a crawler can read your markup without running JavaScript. The Princeton GEO study points the same way: citing sources lifts AI visibility by roughly 40 per cent, adding statistics by around 37 per cent, and quotations by about 30 per cent, while keyword stuffing lowers it. For the source behind each check, see where-every-check-comes-from.