Citation rate vs share of voice: what LLMs actually measure
Citation rate and share of voice do not measure the same thing. The nuance determines what a brand can steer in GEO and what it inherits.
1. Two metrics, two different questions
Citation rate answers a simple question: on a given prompt panel, in what percentage of LLM answers is the brand cited by name, with a link or textual mention? Share of voice answers another question: across all citations delivered by LLMs in a category, what share does the brand capture against named competitors?
The two measures are complementary but they tell different stories. A brand can have a high citation rate on its own branded prompts while running a low share of voice in its category. That is typically the profile of a recognised brand eclipsed by editorial leaders.
Citation rate measures absolute presence. Share of voice measures relative share. A brand steers one, inherits the other, and conflates them at its own expense.
2. What LLMs actually count
No LLM publishes its selection logic. PROEMA internal regressions on 800 Perplexity responses and 400 ChatGPT responses in May 2026 isolate five primary triggers: source authority, content freshness, parsing structure, third-party cross-citations, Schema.org density.
Those five triggers count for citation rate. Share of voice adds a sixth: competitive pressure, namely how many other brands target the same category prompts. On Brussels specialty coffee mid-2026 the pressure is low, so a GEO-native brand quickly captures 15 to 30% share of voice. On Swiss luxury watches the pressure is saturated, share of voice caps under 5% even with a decent citation rate.
3. How each metric is moved
Citation rate moves with Schema density, dateModified freshness, author authority (Wikidata, tier-1 press), and quality of extractive parsing (40 to 80 word paragraphs, clear lists). These are the levers documented in the PGSM v1.0 standard.
Share of voice moves with category coverage. An 800-question FAQ on a vertical captures more share of voice than a 100-question FAQ on the same vertical, at equal authority. It is mechanical: the wider the prompt surface, the higher the probability of appearing in any sectoral answer.
| Metric | Question asked | Primary lever | Typical ceiling |
|---|---|---|---|
| Citation rate | Am I cited? | Schema + authority + freshness | 60 to 80% |
| Share of voice | What share do I capture? | Category editorial density | 30 to 45% |
4. When to steer which
A brand starting its GEO presence steers citation rate first on 30 priority branded and category prompts. The 90-day target is 20 to 30% citation rate on that panel. Enough to prove channel value internally and unlock the second investment round.
A brand installed for six months steers share of voice. Citation rate is now baseline, the strategic question becomes: how to claw points from competitors who are catching up? The answer goes through wider category coverage, tier-1 press, and named-entity defence in Wikidata.
5. The single-dashboard trap
Third-party tools (MentionLab, Profound, Peec) default to a single composite indicator. Convenient for a board readout, misleading for operational steering. The same composite visibility score can hide a 40% branded citation rate and a 5% category share of voice, or the inverse. Strategic implications differ completely.
The Proema Insight standard displays both metrics separately, computed on two distinct panels: a brand-led panel (questions where the brand should be cited by default) and a category-led panel (questions where the brand competes with peers). That separation enables steering, not the aggregate.
Citation rate says whether you exist. Share of voice says what you weigh. A brand that ignores the difference spends its effort on the wrong metric.
source: Proema Insight standard · measurement chapter
The PROEMA GEO diagnostic separates both measures natively on a typical 30-prompt panel (20 category, 10 branded). Read in under 15 minutes, actionable recommendations on the spot.