Citation vs Mention vs Link

1. What “being credited” means here — three outcomes, one event

A generative answer can credit you in exactly three non-equivalent ways — or not credit you at all:

• a citation — your content lifted or paraphrased and tied to an attributable reference;
• a mention — your brand, product, or author named in the prose, with no link;
• a link — a clickable source surfaced, not necessarily tied to any text the answer used.

Definition (GEO Wiki working definition): Attribution in a generative answer is decoupled from grounding: being used as a source and being credited as one are separate events — and credit itself comes in three non-equivalent forms: citation, mention, link.

The three forms all play out at step 4 of the answer loop — synthesis & attribution, the moment after grounding when the engine emits, or withholds, a reference.

2. The three, precisely defined

This table is the load-bearing definition:

A single synthesized answer can carry all three at once — labeled here:

"Generative engines decouple grounding from attribution.[1]   ← citation (chip on a lifted claim)
 According to Otterly.AI, mentions and citations are tracked   ← mention (named, no link)
 as separate KPIs. For more on answer mechanics, see the
 sources below.

 Sources:  [1] geo.wiki/citation-vs-mention
           ▸ example.com/unrelated-page                        ← link (in tray, no matching sentence)"

The canonical practitioner disambiguation, three one-liners you will use constantly:

• “It used my facts, no link or name” → uncredited — grounded ≠ credited (see §3).
• “It named us but we got no traffic” → a mention, not a citation — a different win.
• “It linked us but didn’t quote us” → a link, not a citation — the weakest outcome.

The metric formulae — citation share, share of voice, referral attribution — sit in GEO Metrics; how to earn a mention is the subject of Brand Mentions.

3. Grounded ≠ credited — why attribution is its own gate

The claim that justifies a standalone entry: an engine can ground its answer on your content and still emit zero credit, or name you with no link, or link you without quoting you. Use and credit are decoupled by design, not by bug.

  grounded subset
        │
        ▼
  ┌──────────────────────────────┐
  │  SYNTHESIS & ATTRIBUTION     │
  │  emit credit?                │
  └──────────────────────────────┘
        │
        ├──► citation   (used + credited + reference)
        ├──► mention    (named, no link)
        ├──► link        (URL surfaced, maybe not even used)
        └──► nothing     (used, never credited)

The same grounded passage can exit as any of four outcomes. Most “it used my content and gave me nothing” losses happen exactly here — downstream of grounding, and nothing about being groundable guarantees credit.

Sequence matters. Attribution sits after groundability (Citability — be selectable at all), which sits after retrievability (AI Crawlers — be a candidate at all). An upstream miss makes credit unreachable, so diagnose in loop order; for the full per-step failure map, see Answer Loop §4.

The decoupling is not just conceptual — it is API-visible. Gemini returns groundingChunks (the sources used) separately from groundingSupports (which answer spans are actually attributed back), so “used” and “credited” are literally different fields in the response (Grounding with Google Search). Anthropic’s web search tool makes the same seam concrete: each result carries its own url and cited_text (Web search tool).

4. What the evidence says about attribution honesty — and what it does not

Liu, Zhang & Liang, Evaluating Verifiability in Generative Search Engines (Findings of EMNLP 2023), audited Bing Chat, NeevaAI, Perplexity.ai, and YouChat. The core measurements: on average only 51.5% of generated sentences are fully supported by their citations (citation recall), and only 74.5% of citations actually support their associated sentence (citation precision). The authors call these figures “concerningly low for systems that may serve as a primary tool for information-seeking users.” Being credited is not proof you were used; being used is no promise you will be credited — and even emitted citations are frequently wrong.

The contrast that closes the site’s open loop: Aggarwal et al. measures visibility / impression — being used — and does not measure being credited. That gap is exactly what this entry, via Liu et al., fills. For the critique of Aggarwal’s headline “up to 40%” lift, see the paper entry (arXiv:2311.09735 · ACM DL).

5. Why the distinction is load-bearing for GEO

The “so what”: each outcome maps to a different metric, a different lever, and a different business value. Conflating them mismeasures the whole program.

This separation is industry-standard, not a GEO Wiki invention: Otterly.AI’s KPI taxonomy defines Brand Mentions, Domain Citations, and Share of Voice as three distinct metrics with separate formulas (see Brand Report KPI Definition) — confirming the field already measures “is the brand named” apart from “is the domain cited.”

The load-bearing line: a mention with no click is not a failed citation — it is a different, often slower-compounding win; chasing only link-bearing citations under-counts the entity-prior payoff. The reciprocal of citability’s “necessary, not sufficient”: credit is plural, and each kind pays out differently.

6. How attribution varies by surface (invariant vs delta)

The triad is invariant — citation, mention, and link are distinct everywhere. What varies is density and default form.

Attribution density is not language-invariant — a multilingual-GEO concern.

7. Anti-patterns — misreading the three

This is the entry most likely to be misinterpreted in reporting. Each row: the misread, why it looks right, why it is wrong.

The load-bearing line: you cannot optimize for credit directly — you optimize the gate before it (groundability) and the off-site prior (mentions); attribution is downstream of both. The doing belongs to the playbooks, not to this concept.

8. Why this matters for GEO + how to act

Credit is the payout of the whole answer loop — but it is plural, decoupled, and unreliable, so it must be measured as three things, not one. This entry is the concept; the doing is the playbook.

References

Academic:

• Liu, N. F., Zhang, T. & Liang, P. (2023). Evaluating Verifiability in Generative Search Engines. Findings of EMNLP 2023. arXiv:2304.09848
• Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K. & Deshpande, A. (2024). GEO: Generative Engine Optimization. KDD ‘24. arXiv:2311.09735 · ACM DL · paper summary

Industry / tooling:

• Otterly.AI — Definition of Brand Report KPIs

Official platform documentation (as of 2026-05):

• Google — Grounding with Google Search (Gemini API) · Google Search Central — AI features and your website
• Anthropic — Web search tool
• OpenAI — ChatGPT search
• Perplexity — What is an answer engine, and how does Perplexity work as one?