How the Rubric score works
Rubric is an AI-search citability auditor. It crawls your site and scores how ready your pages are to be quoted when ChatGPT, Perplexity, Google's AI answers and the rest respond to a question. Tracking tools ask an AI questions and count what comes back. Rubric reads the signals the engines read and checks them against a fixed rulebook, so the score comes from measurement, not from a chatbot's reply on the day.
This page sets out how the number is built, so you can check our working.
44 checks, three pillars
Every page is scored against 44 checks. Each check belongs to one of three pillars, and each pillar answers a plain question about whether an engine can use your page.
Known: can an engine attribute you?
Can an engine tell what the page is and who it belongs to? The checks cover Organization and Person schema, sameAs links, content-type schema, schema completeness and validity, canonical tags, internal links, entity density, and titles and meta that aren't duplicated across the site.
Findable: can it lift an answer?
Can an engine reach the page and pull a self-contained answer out of it? The checks cover answer-first openers of 40 to 70 words, question-shaped and claim-shaped headings, sections that open with a direct answer, a single H1 inside the snippet window, titles and meta of the right length, tables and lists, FAQ and HowTo schema, image alt text, an XML sitemap, AI crawlers that aren't blocked at the network, fast server response, and pages that aren't set to noindex.
Trusted: will it quote you?
Does the page give an engine enough reason to stake an answer on it? The checks cover substantive content, statistic density, external source links, a named author, statistics that carry a source, freshness (updated within the last 12 months), and review or rating schema.
Rubric also checks for an llms.txt file, but only as information. It sits outside the 44 checks and never changes the score.
Six engines, each weighted differently
Engines don't value the same signals equally. Rubric weights the 44 checks separately for six of them: ChatGPT, Perplexity, Google AI Overviews, Gemini, Microsoft Copilot and Claude. A check that matters more to one engine carries more weight for that engine, so the report shows how the same page reads to each one. These per-engine fits are modelled from the checks. They aren't observed from the engines themselves.
Grok isn't scored. Rubric shows an advisory note for it instead, because part of what it draws on sits off your site, where an on-page audit can't see.
Render parity: what the engine actually receives
Some engines run JavaScript when they fetch a page and some don't. In Vercel's analysis of AI crawler traffic (December 2024), the crawlers from OpenAI, Anthropic and Perplexity didn't render JavaScript. Gemini, which uses Googlebot's infrastructure, did. Bing says its crawler renders pages with the latest stable version of Microsoft Edge, and Copilot draws on Bing's index.
So schema that only appears after JavaScript runs can look fine in your browser and still be missing for some of the engines you care about. Rubric compares each page's raw HTML with a rendered copy, and scores whether your schema is readable without JavaScript.
The site score is the average of your pages
Rubric scores each page from 0 to 100, then averages the pages that loaded properly. Error pages and tag or category archives are left out, because they have nothing to quote. An average can flatter a site that has a few strong pages, so the report lists every page above and below the line. Read those before the headline.
If your site blocks AI crawlers, through robots.txt or a firewall, the headline score is capped at 25 until the block is fixed. A page an engine can't reach can't be quoted, however good it is.
70 is the line. A page at 70 or above is treated as quotable. Below 70, it needs work.
The same rules, every time
Rubric doesn't ask a model for its opinion, so there's no sampling noise. The same page scored twice under the same rules gets the same score. When we crawled the same 42 agency sites twice, a day apart, in October 2026, the median site moved 0 points.
Every change to the rules gets a new scoring version, recorded with every report. The current version is 2026-10-01. A score means the same thing as any other score from the same version, which is what lets you fix a page, re-crawl it and read the change as a real change.
What the score is, and what it isn't
The score predicts how citable your pages are from the evidence on the page: the content, the structure and the technical set-up. It doesn't count the citations you've already earned, and it isn't a ranking guarantee. We won't tell you a fix buys a specific citation or a traffic lift. Rubric shows you where your pages stand and what to change first. The engines still decide.