---
"@context": "https://schema.org"
"@type": "TechArticle"
"@id": "https://dasarakushi.com/notes/rank-vs-citation-vs-grounding"
headline: "Rank vs citation vs grounding — an AI visibility framework"
description: "Why ranking, AI citation, and model grounding are different systems. A practical framework for measuring AI visibility beyond SERP rank."
author: { "@type": "Person", name: "Dasara Kushi", url: "https://dasarakushi.com" }
datePublished: "2026-05-05"
dateModified: "2026-05-05"
keywords: "AI visibility, Generative Engine Optimization, GEO, AI citations, AI Overviews, grounding, retrieval eligibility"
_format: "text/markdown"
_canonical: "https://dasarakushi.com/notes/rank-vs-citation-vs-grounding"
---

# Rank vs citation vs grounding

> A practical AI visibility framework. Ranking gets you discovered. Evidence gets you cited.

Treating AI visibility as a simple extension of rank breaks in production: pages can rank well and get few citations; others rank lower and get cited often for narrow prompts. **Ranking, citation, and grounding are different systems.** Separate them and measure each directly.

## 1. Three distinct surfaces

- **Rank** — position in traditional SERPs.
- **Citation** — whether your domain appears in the AI answer's source list.
- **Grounding** — whether your content was used as evidence in answer construction, even if not explicitly cited.

The common failure is optimizing one layer and expecting movement in all three.

## 2. Why fan-out is incomplete

Fan-out (expanding a prompt into related sub-queries) explains candidate *discovery* only — not which evidence is chosen or attributed. A better model is four stages:

1. **Retrieval eligibility** — can your page enter the candidate set?
2. **Evidence fitness** — does it contain extractable, query-matching evidence blocks?
3. **Attribution likelihood** — is the source likely to be shown in citations?
4. **Answer assembly constraints** — slot limits, deduping, diversity, UI truncation.

Fan-out is recall plumbing. Citation is evidence selection plus attribution policy.

## 3. Measurement framework

```
Citation Rate        = cited answers / tested prompts
Self Citation Share  = your-domain citations / all citations
Unique Source Share  = unique cited domains / all citations
Source Drift         = prompts where cited domain changes across runs

Grounding Quality (manual)
  2 = directly supported by snippet
  1 = partially supported
  0 = unsupported or inferred
```

Add one binary per page for that exact prompt: **evidence-fit** = High / Medium / Low. It drives most interpretation.

## 4. Example failure mode

A page can be retrieval-eligible yet fail citation: directory/aggregator pages that rank for address queries but don't expose directly citable name mappings in accessible content. The model retrieves the candidate, finds low evidence-fit, then cites alternative domains or returns a sparse answer. That's not model inconsistency — it's evidence quality working as it should.

## 5. Minimal experiment design

```
Prompt Set: 20-30 prompts
Models: 2 (e.g., Gemini + ChatGPT)
Modes: neutral prompt + source-constrained prompt
Runs: 3 per prompt/mode/model
Output per run: answer text, citations, exact supporting snippets, grounding-quality label
```

## 6. Optimization playbook

- **Findability** — crawl/index health, entity clarity, prompt-aligned titles.
- **Extractability** — one claim per paragraph, clear headings, concise answer blocks.
- **Attributability** — verifiable facts, explicit sourcing, stable canonical source URL.

Most teams over-index on findability and under-invest in extractability. Citation performance is usually lost in that gap.

## Closing

For AI visibility, rank is a leading indicator, not the target metric. The target is usable evidence. **Ranking gets you discovered. Evidence gets you cited.** Separating rank, citation, and grounding is the fastest way from SEO folklore to repeatable GEO engineering.

*By Dasara Kushi · May 5, 2026 · https://dasarakushi.com/notes/rank-vs-citation-vs-grounding*