Definition and distinction
Citation rate counts formal source references only: the domain appears as a cited source of a generative answer. That distinguishes it from brand mention density, which counts every mention — including uncited ones. A brand can have high mention density and low citation rate at the same time: the model knows it from training data but does not link it as a primary source.
Both figures belong side by side in reporting because they yield different diagnoses: low mentions with low citations point to an association problem (co-occurrence, entity); high mentions with low citations point to a source problem (citability, crawler access, passage structure).
Measurement
Measurement runs over a curated prompt set (in advisory practice 500 to 2,000 prompts per week), executed across several models and several runs. Citation rate = citing answers ÷ total answers, reported per model and as a weighted total. For robust values: identical prompts over time, documented model versions, multiple runs against answer variance, and a fixed adjudication scheme for edge cases (e.g. a citation without a link).
Benchmarks from advisory practice
Realistic orders of magnitude for domain authorities on subject-matter queries (working values from ongoing measurement, not guarantees):
- Perplexity: 2–8% — the most citation-friendly system, many sources per answer.
- ChatGPT with Search: 1–5% — fewer source slots, stronger consolidation.
- Google AI Overviews: 0.5–3% for top domains — smallest source selection, highest competition.
Absolute values matter less than the trend within your own prompt set and the comparison against direct competitors (share of model).
Drivers and blockers
Citations are favoured by: current, machine-readable publication dates (ISO 8601 in schema), a clear author entity, deduplicated canonicals, high domain authority and passage-level structured content — H2-led answers, lists, FAQ blocks (see passage ranking and the QUEST heuristic).
Citations are blocked by: content behind JavaScript rendering or login gates, unclear publication dates, blocked AI crawlers, and passages that only make sense in the context of the whole page.
Related terms
Citation rate is one of the seven primary metrics in the LLM Citation Benchmark. It sits alongside share of model, PVI and brand mention density. How citations can be optimised per model is covered in Claude citation optimisation.
Citations are countable — and therefore steerable
Citation rate translates the vague “do we appear in AI answers?” into a countable event with baseline, trend and competitive comparison. It belongs in every GEO report as the headline KPI.
FAQ on citation rate
What is a good citation rate? ▾
Context-dependent. Working values from ongoing measurement for domain authorities on subject-matter queries: Perplexity 2–8%, ChatGPT with Search 1–5%, Google AI Overviews 0.5–3%. Trend and competitive comparison within the same prompt set matter more than the absolute value.
How do citation rate and brand mention density differ? ▾
Mention density counts every brand mention in answers — even without a source reference. Citation rate counts only formal citations with source attribution. The combination shows whether you have an association problem or a source problem.
How many prompts does a robust measurement need? ▾
In practice, 500 to 2,000 prompts per week across several models and several runs have proven reliable. Smaller sets show tendencies but vary strongly; single prompts are worthless as measurement.
Can citation rate be converted directly into traffic? ▾
No. A citation proves source presence, not clicks. The relationship with traffic depends on the system (link display), the prompt type and user behaviour — converting it into visitors or revenue would be invented.