Skip to main content
CiteCheckup logoCiteCheckup

Scoring methodology

How CiteCheckup scores a page

CiteCheckup evaluates information it can retrieve from one public page. The result explains what to fix; it does not predict whether an AI system will cite the page.

100 points total

Four weighted dimensions

25 / 25 / 35 / 15

  1. 25

    Crawl and index access

    Can public crawlers retrieve the page, follow its controls, and identify the canonical, indexable content?

  2. 25

    Machine-readable structure

    Do metadata, language declarations, headings, structured data, lists, tables, and authorship signals describe the page clearly?

  3. 35

    Citation-ready content

    Does the page provide direct answers, useful topic coverage, evidence, sources, comparisons, steps, summaries, and definitions?

  4. 15

    Brand and trust

    Are the brand, responsible people or publisher, entity details, freshness, and contact paths clear and consistent?

Every rule area in the audit

These 30 checks make up the four category scores. The report lists each one, so you can see where every point came from.

Crawl and index access

25 pts
  • HTTP response
  • robots access
  • AI crawler access
  • noindex
  • canonical URL
  • sitemap discovery
  • server-rendered body

Machine-readable structure

25 pts
  • page title
  • meta description
  • declared page language
  • heading hierarchy
  • structured data
  • lists
  • tables
  • FAQ structure
  • author and date metadata

Content and sources

35 pts
  • direct answer
  • topic coverage
  • question and answer coverage
  • facts and numbers
  • source links
  • comparison support
  • actionable steps
  • summary
  • definitions

Brand and trust

15 pts
  • brand consistency
  • entity schema
  • author or publisher
  • freshness
  • about or contact links

What each status means

The status tells you whether a check passed, needs attention, failed, could not be confirmed, or does not apply. Missing information is never silently counted as a failure.

Pass
The page meets the rule and earns the available points.
Warn
The page partly meets the rule and earns partial points.
Fail
The required item is missing, invalid, or blocked.
Unknown
The audit could not verify the signal with enough confidence. Unknown is not confirmed missing and is excluded from the score denominator.
Not applicable
The rule does not apply to this page type and is excluded from both score and coverage denominators.

Score and evidence coverage

Score = known earned points / known maximum points x 100. Only Pass, Warn, and Fail checks with sufficient evidence are known.

Coverage = known maximum points / applicable maximum points x 100. Coverage measures how much of the applicable audit could be verified, not page quality.

When coverage is below 80%, the score is provisional. Fixing access or evidence gaps may change the score even when the visible copy does not change.

What each result is based on

Each check shows the public source, what was found, and how certain the result is. You can review the page detail behind the score.

Unknown means the checker could not confirm the item. Fail means it confirmed that the requirement was missing, invalid, or blocked. The report keeps those cases separate.

How top actions are ranked

Top actions include only Warn and Fail checks. Priority = lost points x confidence / effort. Lost points are the check maximum minus earned points; effort uses low, medium, or high divisors.

Items that can recover more points with less work appear first. Impact and the fixed rule order break ties. This is a suggested editing order, not a promise of traffic or citations.

llms.txt is a non-scoring hint

A public llms.txt file can provide additional context in the report, but llms.txt remains a non-scoring hint. Its presence or absence does not change score or coverage.

Scope

Supported languages and page types

English, Chinese, and Japanese are first-class for language detection and content rules. The audit is designed for public HTML pages such as landing, product, service, article, guide, comparison, and reference pages. Mixed or unsupported language is labeled, and page-type-specific checks can be Not applicable.

Limits of a public page audit

The audit can assess only public signals it can observe for the submitted URL. Login walls, access restrictions, interaction-only content, incomplete metadata, and a changing live page can reduce evidence coverage. One report is a point-in-time page audit, not a site-wide assessment.

ChatGPT, Claude, Gemini, Perplexity, Copilot, and other AI products use signals and selection logic that a public page audit cannot inspect, so public signals cannot guarantee a citation.

How to read your results

Read the score and coverage together. Then open the individual checks before deciding what to change for your readers.

A higher score means more known rules earned points. It does not prove the page will be discovered, ranked, mentioned, or cited.

Data handling and local report history

The submitted URL and public page data are processed transiently for the requested audit. There is no server report database. Brand and topic are optional advanced overrides and are used only for that audit.

The current browser can retain the last 10 reports, including the URL, scores, evidence, and fixes. Clear removes that local browser history.

Advertisement

Related reading