Citation Rate
Consensus definition
Citation Rate sits one layer above mention: the brand appears as a cited or linked source the answer relies on, not just a name in the text. AI-visibility tools report it widely, but the exact counting rules differ from tool to tool — and since 2026 two platforms report a first-party version of their own: Bing Webmaster Tools counts citations per grounding query and a Citation Share against all cited sites1; Search Console reports impressions in Google's generative AI features, without citations as such2.
rhinegold operator caution
Rhinegold deliberately runs two parallel citation definitions and does not unify them: a mention-with-source reading (a brand mention in the answer text paired with a grounding URL) and a URL-citation reading (body and grounding URLs, de-duplicated per answer). They answer different questions and feed different dashboards. The discipline that makes either trustworthy is span-aware matching and consistent de-duplication — and always stating which definition a number uses. The same prompt-set discipline that protects Mention Rate applies here: a citation rate computed over prompts which already name the brand is closer to a tautology than to competitive evidence, and unaided and brand-aided pools must be reported separately. See Aided vs Unaided Brand Recall in GEO. Public data shows how far the two pools sit apart: across 57 million citations in B2B answers, brand-owned domains took 30.7 % of citations on branded prompts and 2.2 % on unbranded ones3.
Operational use
Citation Rate is the authority layer: when you need to know not just whether the brand is named, but whether the engine treats it as a source worth standing on.
Measurement boundary
It is provider-dependent — grounding exposure varies by engine, so the same brand can look very differently cited across providers. And the two definitions can diverge substantially, so a citation rate is meaningless without saying which one produced it. A citation also has three levels that providers blur: a source retrieved for the answer, a source listed in the provider's source list, and a source tied to a statement in the text — the same page can count under one and not the others. Because the counting unit, the denominator and the provider blend each move the number, a series is only readable with its method frozen — see Measurement Consistency. The same applies to the access path: API and UI configurations of the same provider return materially different source mixes, so the path belongs in the disclosure — see API vs UI Measurement.
What can still be observed
Even where a provider hides its grounding metadata, the body-URL definition stays countable from the answer text itself — so a citation signal remains observable on every provider, just not always the same one.
Distinct from
Against Mention Rate, which is presence without a source. Against Brand Recommendation Share, which weights recommendation by competition. Against Grounding, which is the retrieval mechanism behind a citation, not the visible citation itself. Against Source Authority, which qualifies the citation by the weight of the citing domain — a citation rate is volume, source authority is quality.
Operational note
Common mistakes
- Conflating Citation Rate with Mention Rate — presence is not authority.
- Comparing citation rates computed under different definitions — including a platform's own Citation Share against a prompt-set rate.
- Ignoring that grounding exposure, and thus visible citation, differs across providers.
- Reporting a single aggregate Citation Rate over a prompt set that mixes brand-aided and unaided prompts — see Aided vs Unaided Brand Recall in GEO.
Where consensus is missing
There is no industry-standard definition of a citation in AI answers. Rhinegold runs two on purpose (see the operator position) rather than forcing a single lossy one — so cross-tool citation rates are not directly comparable. Bing's Citation Share adds a third, first-party definition (per grounding query, against all cited sites) that matches neither of the two above.
Sources & deeper reading
- Google — Grounding with Google Search (Gemini API): groundingMetadata returns the sources behind an answer
- 1Bing Webmaster Tools — "New AI Visibility Insights: Intents, Topics, Citation Share, Compare" (16 June 2026; AI Performance report since 10 February 2026: citations and grounding queries across Copilot, Bing AI answers and partner surfaces)
- 2Search Console Help — "Generative AI performance report (Search)" (impressions only, AI Overviews and AI Mode combined; all websites from 31 August 2026)
- rhinegold Insights, Episode 04 — How to Measure What You Can't See (GEO metrics for B2B)
- rhinegold citation-extraction methodology (body ∪ grounding, mother/daughter consolidation) — applied with engaged clients and partners.
Update history
- 2026-10-08 — Added the first-party citation data that now exists (Bing Citation Share, Search Console Generative AI report) as a third definition, the branded/unbranded citation gap from Foundation × AirOps, the retrieved/listed/supported distinction, and the link to Measurement Consistency.