AI Visibility Score Lies: Visibility vs Citability

A site can be highly Visible (every AI engine knows the brand exists) and rarely Cited (engines mention the name but never link to the site). Or Cited but never Recommended (the URL appears in answers but engines do not suggest the site as a top option). Or Recommended but invisible to direct queries (engines suggest the brand for adjacent topics but cannot describe what the brand does when asked directly). chudi.dev measured 53% brand-recognition rate and 0% topic-authority citation rate at one audit point. Same site, same engines, two different problems with two different fixes. The single AI visibility score that most tools in the category ship cannot tell you which problem you have, which is the same as not measuring it at all.

This post defines the three axes, shows real asymmetric data across them, and walks through axis-specific remediation. It composes with The 0% ChatGPT Citation Trap, which covered why a single rate can be wrong on one axis at a time, and Three-Anchor Calibration, which covered how to validate the rate is calibrated. This post is the framework that makes those rates actionable instead of just measurable.

The Single-Score Fallacy#

Most AI visibility tools render one number on the dashboard. HubSpot's AEO Grader gives a letter grade. Semrush's AI Visibility Index ranks brands on a 0-100 scale against competitors. The various rebranded SEMrush wrappers and new-entrant audit tools all converge on the same shape: one big number, optionally broken out by engine.

The single number is easy to render, easy to compare, and easy to put in marketing copy. It is also unactionable. A buyer who sees 47% AI visibility cannot tell whether the gap is in entity authority, content quality, or structural retrievability. Each of those has a different remediation cost and a different timeline. Each requires a different team. Confusing them produces three months of wrong work followed by an audit that says the remediation did not move the number.

The category default is to ship the score because it tests well in product demos. A C-suite buyer wants a number that goes up over time. The single rollup gives them that. The downside is that the number going up does not tell the operator what to do, so the cycle of measure-fix-remeasure stalls when the operator does not know which fix to try.

The right design is to ship V/R/C separately AND a rollup score. The rollup is the dashboard-ready feel-good number. The separation is what makes the score useful. The category will move toward this two-layer rendering once buyers learn to ask for the separation. Tools that lead the shift mint the terminology that everyone else has to adopt.

The Three Axes Defined#

V/R/C decomposes AI visibility into three measurable, independent dimensions.

V (Visibility) is the entity-recognition axis. It answers: when an AI engine encounters the brand name, does the engine know what the brand is? Visibility is measured by querying engines with brand-direct prompts ("what is [brand]", "tell me about [brand]", "what does [brand] do") and counting responses that correctly identify the brand. High Visibility means the entity is recognized. Low Visibility means the engine treats the brand as unknown or mistakes it for something else.

The signals that drive Visibility are mostly external to the site: presence in the model's training corpus, Wikipedia or Wikidata entries, mentions in authoritative third-party sources, sameAs links across surfaces. Schema markup on your own site contributes but rarely dominates: a brand that has never been mentioned in the training corpus will not become Visible just because its homepage has perfect Person and Organization schema. Visibility is the slowest axis to move because training data updates lag by months.

R (Recommendability) is the suggestion axis. It answers: when an AI engine answers a question in your category, does it suggest your brand as one of the options? Recommendability is measured by querying with category prompts that do not name the brand ("best tool for X", "recommend a service for Y", "top alternatives to [competitor]") and counting how often the brand appears in the response. High Recommendability means the engine treats the brand as a credible option. Low Recommendability means the engine answers the question without considering the brand.

The signals that drive Recommendability are content-side: answer-first structure, FAQPage schema with comparative content, freshness signals on category pages, comparative content that positions the brand against alternatives. A brand that is Visible (the engine knows it exists) but does not appear in category responses is failing on the R axis: the engine knows the brand but does not consider it relevant when alternatives are requested. The remediation is content work, not entity work.

C (Citability) is the URL-citation axis. It answers: when an AI engine references the brand in an answer, does it link to the brand's URL as a structured citation? Citability is measured by counting structured citations (clickable links in the response) across all queries, not just brand-direct ones. High Citability means the engine treats the brand's site as a primary source. Low Citability means the engine knows the brand but cannot or will not cite it.

The signals that drive Citability split into access and substance. Access: robots.txt allowing GPTBot / ClaudeBot / PerplexityBot, an llms.txt / ai.txt discovery index, server-side rendering (so crawlers see content without JS execution), and canonical URLs that resolve cleanly. Substance: answer-first, original, extractable content the engine can lift, backed by off-site entity authority. JSON-LD schema (Article, TechArticle, FAQPage) helps engines disambiguate the entity but is not on its own a proven citation lever. Access and rendering fixes are the fastest to take effect: they can make a page eligible for citation within two to four weeks of a re-crawl, though eligibility is not the same as a guaranteed lift.

The three axes are not strictly hierarchical (V > R > C). C can exceed R when an engine cites the brand for one query but recommends a competitor for another. R can exceed V when the engine treats the brand as a category option without quite knowing what the brand does. The independence is what makes the framework load-bearing: each axis can move on its own, and conflating them hides which is which.

Why Independence Matters#

The clearest evidence the axes are independent is the asymmetric data on real sites. A few patterns we have seen:

A single AI visibility score collapses these four diagnostically distinct patterns into "the customer's number is around 30%". The single number is the same; the right next move is different in each row. The framework matters because it surfaces what the single score hides.

Per-Axis Remediation#

Each axis has different drivers, different time-to-impact, and different teams responsible.

V remediation: entity authority work. Author or update Wikipedia and Wikidata entries for the brand if they are missing. Add Person and Organization schema with consistent sameAs arrays across the site, social profiles, GitHub, and any other surface the brand owns. Get cited by authoritative third-party sources in the category. The shortest path is usually a Wikipedia update plus a few high-DR brand mentions; the long-term work is being mentioned often enough that future training cycles include the brand. Time-to-impact: 2-6 months for the next training cycle to absorb new mentions, immediate-to-2-weeks for Knowledge Graph and per-query Knowledge Panel signals.

R remediation: content quality and answer-first structure. Rewrite the most-relevant pages so the direct answer appears in the first 100 words. Add FAQPage schema with clear question-answer pairs. Publish category-defining or comparative content that positions the brand against alternatives. Improve freshness signals (dateModified that reflects real updates, not cosmetic ones). Time-to-impact: 3-6 weeks after the content lands and crawlers re-index.

C remediation: structural retrievability. Make pages reachable and parseable by AI crawlers first: audit robots.txt so GPTBot, ClaudeBot, PerplexityBot, and Google-Extended are not blocked, publish llms.txt and ai.txt at the canonical paths, switch to server-side rendering if the site ships blank HTML and hydrates client-side (most LLM crawlers do not run JS), and verify canonical URLs resolve cleanly. Add JSON-LD schema (Article, TechArticle, FAQPage) to help engines disambiguate the page, not as a citation lever in itself. Time-to-impact: access and rendering fixes can make pages eligible for citation within 2-4 weeks after they ship and crawlers re-visit. Fastest of the three axes to move, though eligibility still depends on answer-first substance and entity authority.

The right priority order is bottleneck-first, weighted by time-to-impact. If V is 80%, R is 60%, C is 10%, the bottleneck is clearly C and structural work moves the score fastest. If V is 10%, R is 5%, C is 5%, the bottleneck is V and no amount of structural work will move the rollup until the entity is recognized at all. The three-axis matrix surfaces the bottleneck; the single rollup score does not.

The V/R/C Audit Format#

A V/R/C audit produces a 3x3 matrix per audit run: three engines (ChatGPT, Perplexity, Claude) by three axes (V/R/C), nine cells, each a rate.

{
  "audit_run_at": "2026-05-18T14:22:33Z",
  "site": "example.com",
  "matrix": {
    "openai_gpt_4_1": {"V": 65, "R": 38, "C": 22},
    "anthropic_claude_sonnet_4_6": {"V": 60, "R": 35, "C": 18},
    "perplexity_sonar": {"V": 70, "R": 42, "C": 28}
  },
  "rollup_avr_score": 42,
  "bottleneck_axis": "C",
  "bottleneck_engine_avg_pct": 23,
  "remediation_priority": [
    "C (citability): structural retrievability",
    "R (recommendability): content quality + answer-first",
    "V (visibility): entity authority"
  ]
}

The rollup_avr_score is the dashboard-ready single number for buyers who want it. The matrix is the actionable data for the operator who needs to fix something. The bottleneck_axis names the lowest-scoring axis explicitly so the next remediation step is obvious. The remediation_priority orders the three axes by impact-cost ratio.

This is the same shape the calibration receipt covered in Three-Anchor Calibration anchors against. Calibration produces a trustworthy rate per cell; V/R/C separation produces actionable rates per axis. They compose: a tool needs both layers to ship audit numbers a buyer can both trust and act on.

What This Means for Tool Buyers and Builders#

For buyers: stop accepting single AI visibility scores as the audit output. Ask vendors for a V/R/C breakdown per engine. The vendors who can produce one are the vendors whose numbers are actionable. The vendors who cannot are shipping a feel-good dashboard, not a remediation roadmap. Make the breakdown a contract requirement before signing any audit engagement.

For builders: ship V/R/C breakdown alongside the rollup, not instead of it. The rollup is the dashboard-ready number that wins the demo; the breakdown is what wins the contract renewal because customers can act on it and see real movement. Tools that lead this shift become the field's vocabulary for the dimension a buyer asks about ("what is my Citability rate?" instead of "what is my AI visibility score?"). Vocabulary leadership is durable competitive advantage in a forming category.

The category is converging on shared methodology this quarter. The three-axis decomposition is the methodology that makes the whole stack (calibration receipts, three-anchor calibration, V/R/C separation) cohere into something a buyer can verify and act on. For the open-source receipt schemas and reference implementations, see the AI Visibility Readiness Framework on GitHub. For the citability.dev audit that ships V/R/C separation as the default output (rollup AVR Score plus per-engine matrix plus calibration receipt), start at citability.dev/assess. For the deeper conceptual context on why entity recognition (the V axis) is the slowest pillar to move and the highest ceiling, Chudi's Entity Mention Graph and AI Visibility on chudi.dev covers the entity-authority side that V/R/C's V axis depends on.

Buyers will eventually ask for the breakdown because the alternative (single scores that do not move under remediation) erodes trust. Tools that lead the shift mint the vocabulary. Tools that follow inherit it.