Citability

1. What citability is

Citability is the property that decides whether an already-retrieved passage is selectable for grounding. It is not about being found — it is about being usable once found.

Definition (GEO Wiki working definition): Citability is the structural property that decides whether a retrieved passage can be lifted, intact, into a generative answer — independent of whether the source is trusted enough to be used at all.

In the taxonomy this is the content-structure signal — the shape half of grounding. Its sibling is E-E-A-T, the trust/authority half. The two are orthogonal levers on the same step; §3 makes the boundary explicit. Hold the one-sentence version for now: structure decides if a passage can be lifted; trust decides if the source is allowed to be.

2. Retrieved ≠ grounded — why citability is its own gate

The claim that justifies a standalone entry: a page can be crawled, indexed, and retrieved into the candidate set — and still never used. Retrieval makes you a candidate; grounding decides which candidates the model is permitted to base the answer on. Citability is the lever for exactly that second gate.

  candidate passage set
        │
        ▼
  ┌─────────────────────────────┐
  │  CITABILITY GATE             │
  │  self-contained?            │  ── no ──►  dropped
  │  answer-shaped?             │             (retrieved,
  │  liftable with attribution? │              never used)
  └─────────────────────────────┘
        │ yes
        ▼
  grounded subset ──► synthesis ──► (maybe) attribution

The gate is where most “found but not cited” losses happen. A page that is retrieved but whose passages are not liftable is dropped here — and nothing downstream can rescue it.

Sequence matters. Citability sits after retrievability (AI Crawlers — be a candidate at all) and before attribution (Citation vs Mention — be credited once used). An upstream miss makes this lever irrelevant, so diagnose in loop order; the full per-step failure map sits in Answer Loop §4.

3. Citability vs E-E-A-T — the two orthogonal grounding levers

The single highest-value disambiguation on this page. Both gate grounding; they are independent. Most “I did everything and still wasn’t used” confusion comes from collapsing these two into one.

The load-bearing line: a trusted source written as a wall of text still loses grounding; a perfectly chunked page with no authority loses on trust. You need both — they do not substitute. Trust signals — author credentials, first-hand experience, citation density as a quality signal — sit on the E-E-A-T axis.

4. The anatomy of a citable passage

This is the load-bearing taxonomy of citable structure. The same seven signals are operationalized in the Citability playbook:

Microsoft states the structural case plainly: “Clear headings, tables, and FAQ sections help surface key information and make content easier for AI systems to reference accurately” (see Bing Webmaster — AI Performance).

4.1 Self-contained chunk

The atomic unit. If a passage needs the paragraph above it to make sense, the engine cannot lift one cleanly.

• ✓ “Citability sits after retrieval and before attribution in the answer loop.”
• ✗ “As noted above, it sits between those two — see the earlier diagram.”

4.2 Direct-answer / TL;DR block

State the answer, then justify it. Inverted-pyramid passages are selected more often because the liftable claim is at the top.

• ✓ “GEO is not SEO relabelled. Keyword stuffing did not raise AI-answer visibility; content substance did.”
• ✗ Three paragraphs of context before the claim appears.

4.3 Q&A / FAQ structure

Question-shaped headings line up with the sub-queries fan-out produces (see Answer Loop §3.1). Match real questions, not invented ones.

• ✓ ### My page was retrieved but not cited — why?
• ✗ ### Considerations regarding retrieval dynamics

4.4 Step / HowTo structure

An ordered procedure lifts as one clean unit. Numbered, imperative, one action per step.

• ✓ A numbered list the engine can quote whole.
• ✗ “First you should consider… and then it may be worth…” prose.

4.5 Citable table / list

Each row must be readable alone — caption it, label its columns, no row that depends on the prose around it.

• ✓ This page’s §4 table: every row stands by itself.
• ✗ A table whose rows mean nothing without the paragraph before it.

4.6 Heading-hierarchy discipline

Clean nesting makes passages addressable; engines retrieve and quote by section. No skipped levels, no decorative headings.

• ✓ H2 → H3 → H3, each a real unit.
• ✗ H2 → H4, or headings used purely for visual size.

4.7 Liftable quotable sentence

The smallest unit of citability: a single claim that survives extraction with attribution intact. Crisp beats hedged.

• ✓ “Retrieval makes you a candidate; grounding decides if you are used.”
• ✗ “It could perhaps be argued that, in some cases, retrieval may not always lead to use.”

5. What the evidence actually says — and what it does not

The empirical anchor, carried with the same honesty discipline as the source paper. Aggarwal et al. tested nine content rewrites: content-substance rewrites — cite sources, add statistics, add quotations — measurably raised answer visibility, while Keyword Stuffing, the SEO reflex, did not (and could hurt). This is early evidence that GEO is not SEO tactics relabelled.

The position, stated plainly: take the direction, discard the number as a planning input. The single-actor lift is an upper bound, not the equilibrium once competitors optimize the same engine (C-SEO Bench, Puerto et al., NeurIPS ‘25 D&B). For the full critique, see the paper entry.

One adjacent line, routed not expanded: an engine can ground on your text and still not credit it — verifiability and attribution are a separate problem, governed by Citation vs Mention, not by citability (Liu et al.).

6. Anti-patterns — when “optimizing for citability” backfires

Citability is the entry most likely to be over-applied. Each anti-pattern below looks like the signal it imitates and fails because it trips a trust or AI-spam filter.

The load-bearing line: citability is necessary, not sufficient. Perfect structure cannot rescue a thin or untrustworthy page — that gap is E-E-A-T’s. And structure without substance is detectable: over-optimized, low-value patterning is itself a penalized signal, covered by AI Content Detection. Google’s own guidance is that there are “no special optimizations necessary” beyond helpful, original content (see AI features and your website; Succeeding in AI search).

7. How citability varies by surface (invariant vs delta)

The structural property is invariant — self-contained, answer-shaped, liftable wins everywhere. What varies is chunk-size tolerance and which answer-block shape each surface rewards.

Two routed lines, not expanded here: being citable presupposes being fetchable at all — OpenAI is explicit that appearing and being cited requires not blocking its search crawler (see Publishers and Developers FAQ), which is AI Crawlers’ domain. And chunk/answer-block citability is not language-invariant in practice — that is Multilingual GEO’s.

8. Why this matters for GEO + how to act

Grounding is the choke point the answer loop §3.3 calls “the highest-leverage step for most practitioners” — and citability is the lever for it. This entry is the concept; the doing is the playbook. Structure is not a checklist floating in the abstract: each of the seven signals is a pin on the grounding step.

References

Academic:

• 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
• Puerto, H., Gubri, M., Green, C., Oh, S. J. & Yun, S. (2025). C-SEO Bench: Does Conversational SEO Work? NeurIPS ‘25 Datasets & Benchmarks. arXiv:2506.11097
• Liu, N. F., Zhang, T. & Liang, P. (2023). Evaluating Verifiability in Generative Search Engines. Findings of EMNLP 2023. arXiv:2304.09848

Official platform documentation (as of 2026-05):

• Google Search Central — AI features and your website · Succeeding in AI search
• Microsoft Bing — Introducing AI Performance in Bing Webmaster Tools (Public Preview)
• OpenAI — ChatGPT search · Publishers and Developers FAQ
• Perplexity — What is an answer engine, and how does Perplexity work as one?