Methodology · public trust surface

Every AVR score must be reconstructable from receipts.

citability.dev treats AI-answer visibility as an evidence problem. Scores summarize model outputs, but every score must drill down to the raw run, model version, prompt, extracted citations, classification logic, and hash.

01 /V · R · C axes

VVisibility

Whether the brand appears at all for the query and model.

RRecommendability

Whether the answer frames the brand as a viable recommendation.

CCitability

Whether the answer links or cites the brand/domain as a source.

02 /Scoring model

AVR = weighted(V, R, C)

V = visibility across query × model receipts
R = recommendation quality and rank position
C = citation/link extraction quality

band: ≥70 healthy · 40-69 at-risk · <40 invisible

03 /Proof — valid audits

Every score ships with a receipt.

A receipt is immutable, timestamped, model-specific, query-specific, hash-addressed, and suitable for export, appeal, and public verification.

Open sample public receipt

GET /v1/receipts/rcpt_01HXK9F2

{
  "id": "rcpt_01HXK9F2",
  "query": "best b2b saas payment processor",
  "verdict": "mixed",
  "avr": 62,
  "models": ["chatgpt/gpt-5.1", "claude/4.5", "perplexity/sonar", "gemini/2.5"],
  "signature": "sha256:a14f...b902"
}

04 /Data sources

OpenAI

ChatGPT

gpt-5.1

Anthropic

Claude

claude-4.5-sonnet

Perplexity

sonar-pro

Google

Gemini

gemini-2.5-pro

05 /Observable pipeline

Normalize project, domain, models, query set, and schedule.

Dispatch prompt jobs across pinned model versions.

Persist raw responses, timing, sources, and provider metadata.

Extract linked citations, in-text mentions, and referenced domains.

Classify cited, mentioned, competitor-mentioned, or absent.

Calculate V/R/C components and AVR rollup.

Sign receipts and expose public verification when enabled.