Summarize this article with AI
Ask with ChatGPTOpen with this article
Ask with ClaudeOpen with this article
Ask with PerplexityOpen with this articleThe vocabulary is broken. Three teams in the same $40M ARR B2B SaaS will report three different citation share numbers in the same week, none of them wrong, all of them measuring something different. One team counts mentions. Another counts linked citations. The third weights by prompt volume. The CMO concludes the data is unreliable. The data is fine. The definitions are not.
This is the depth-first reference glossary the SERP does not ship. Fifteen GEO metrics, one structured block each, every definition written in citation-grade form (under 40 words, the shape an LLM extracts cleanly). Every metric carries a formula, a unit, a range, a when-to-use note and the metric most teams confuse it with. Bookmark the page, link to it from your reports, end the vocabulary fight before the next QBR.
Citation metrics
The four metrics that count linked citations: how often, how competitive, how diverse, and how much of your visibility is unlinked.
Citation Rate
The percentage of monitored prompts in which an AI engine cites your domain via a clickable source link, measured per engine and aggregated across the prompt panel.
- Formula
- (prompts citing your domain / total prompts in panel) × 100
- Unit / range
- Percent, 0 to 100. Competitive median 8 to 15%, top quartile 30%+
- When to use
- Weekly trend on a fixed prompt panel. The closest GEO analog to indexed and ranking.
- Confused with
- Citation Share (which is competitive, not absolute)
Citation Share
Your domain's slice of all citations awarded across the prompt panel, capturing competitive position rather than absolute coverage of the panel.
- Formula
- (your citations / total citations across all domains) × 100
- Unit / range
- Percent, 0 to 100. Healthy above 12% in a 5-competitor category.
- When to use
- When the question is competitor activity, not your absolute reach.
- Confused with
- Share of Voice (which counts mentions, not citations)
Citation Source Diversity
The count of distinct URLs from your domain cited at least once across the prompt panel, divided by the total URLs in your sitemap. Measures depth of indexed surface, not just hero pages.
- Formula
- unique URLs cited / total URLs in sitemap
- Unit / range
- Ratio, 0 to 1. Healthy above 0.15 for content-led sites.
- When to use
- Catching monoculture risk where one page carries the brand.
- Confused with
- Citation Rate (volume, not surface area)
Ghost Citation Rate
The percentage of AI answers that mention your brand by name without linking to your domain. Visibility you earn but do not measure in referral logs.
- Formula
- (unlinked mentions / total mentions) × 100
- Unit / range
- Percent, 0 to 100. Typical 40 to 60%, most mentions are ghost.
- When to use
- Reconciling brand-monitoring counts with GA4 referral data.
- Confused with
- Mention Rate (which counts all mentions, ghost rate isolates the unlinked share)
Mention and sentiment metrics
The three metrics that capture brand presence and tone, even when no link is involved.
Mention Rate
Percentage of prompts that produce any reference to your brand, linked or unlinked, regardless of position or sentiment. The broadest top-of-funnel signal.
- Formula
- (prompts mentioning brand / total prompts) × 100
- Unit / range
- Percent, 0 to 100. Often 2 to 4× higher than citation rate.
- When to use
- Top-of-funnel awareness. Mentions typically precede citations by 2 to 4 weeks.
- Confused with
- Citation Rate (links required, not just name)
Mention Frequency
The average count of brand references per response in which your brand appears at least once. Measures depth of mention within an answer, not breadth across answers.
- Formula
- total mentions / responses with at least 1 mention
- Unit / range
- Count, typically 1.0 to 5.0+
- When to use
- Detecting whether the model treats your brand as a primary subject or an afterthought.
- Confused with
- Mention Rate (breadth across prompts, not depth within)
Sentiment Score
Aggregated polarity of language surrounding brand mentions in AI answers, scored per occurrence on a −1 to +1 scale and averaged across the panel.
- Formula
- Σ(polarity per mention) / N
- Unit / range
- −1.0 to +1.0. Healthy above +0.3, flag below 0.
- When to use
- Monthly. Pair with sentiment ratio (% positive minus % negative) for context.
- Confused with
- Sentiment Ratio (which counts buckets, not polarity scores)
Coverage and competitive metrics
The three metrics for measuring presence at scale and category authority.
Share of Voice (SoV)
Your share of all brand mentions, linked or unlinked, inside AI answers across a defined prompt set, weighted equally per prompt regardless of search volume.
- Formula
- (your mentions / total brand mentions in panel) × 100
- Unit / range
- Percent, 0 to 100. Top-quartile B2B SaaS 18 to 25%.
- When to use
- Narrative presence and brand recall. Broader than citations.
- Confused with
- Share of Model (which is volume-weighted by prompt frequency)
Share of Model (SoM)
Prompt-volume-weighted share of inclusion in AI answers, where each prompt is weighted by its real-world search frequency rather than counted equally.
- Formula
- Σ(inclusion × prompt volume) / Σ(prompt volume)
- Unit / range
- Percent, 0 to 100
- When to use
- When prompt panels mix high and low traffic queries. The truer commercial signal.
- Confused with
- Share of Voice (which is unweighted)
Branded vs Unbranded Ratio
Split of your citations between prompts that name your brand explicitly and prompts that do not. Unbranded citation share is the harder, more valuable share for category authority.
- Formula
- branded citations / unbranded citations
- Unit / range
- Ratio, 0.2 to 5.0+. Below 1.0 indicates strong category authority.
- When to use
- Distinguishing brand strength from category visibility.
- Confused with
- Mention Rate (which does not split by intent)
Position and prominence metrics
Two metrics for the same intuition at different granularities: where in an answer your brand actually appears.
Passage Prominence
A position-weighted score of where your citation appears within an AI answer, capturing recommendation strength. Lead position carries 3, body 2, trailing source list 1.
- Formula
- Σ(weight × cited) / N, weights = {3, 2, 1, 0}
- Unit / range
- 0 to 3.0
- When to use
- When two brands have equal citation rate but different perceived authority.
- Confused with
- Position Prominence (sentence-level, not block-level)
Position Prominence
The sentence-level position of the first occurrence of your brand within a response, normalized from 0 (first sentence) to 1 (last). Below 0.25 is lead-position inclusion.
- Formula
- (sentence index of first mention) / total sentences
- Unit / range
- 0 to 1.0. Lead-position threshold is below 0.25.
- When to use
- A/B testing rewrites for lead-position lift. More granular than passage prominence.
- Confused with
- Passage Prominence (paragraph block weight, not sentence index)
Engine and resilience metrics
The three metrics that measure how well your visibility is distributed across engines and how quickly it bounces back from a shock.
Engine Spread
Standard deviation of your citation rate across the six major engines (ChatGPT, Claude, Gemini, Perplexity, Google AI Mode, Copilot). Low values mean balanced presence, high values mean concentration risk.
- Formula
- σ(citation rate per engine) for e in {6 engines}
- Unit / range
- Percentage points, 0 to 50. Below 8 = healthy, above 15 = concentrated.
- When to use
- Detecting over-reliance on one engine before that engine patches you out.
- Confused with
- Cross-engine coverage (count of engines, not variance)
Answer Churn
Week-over-week percentage of monitored prompts whose top-cited sources change for the same query. Measures GEO volatility and the impact of model updates.
- Formula
- (prompts with changed top-3 citations / total prompts) × 100
- Unit / range
- Percent, 0 to 100. Baseline 15 to 25%, post-update spikes to 50%+.
- When to use
- Detecting algorithm shifts and rewrite urgency. Action threshold above 35% per week.
- Confused with
- Recovery Half-Life (time to rebound, not rate of change)
Recovery Half-Life
The median number of days for a paragraph rewrite to recapture lost citation share after a model update or competitor displacement, measured to 50% of pre-event baseline.
- Formula
- days until citation share returns to 50% of pre-event baseline
- Unit / range
- Days, 5 to 45. Manual workflows 30 to 45. Tools that close the measure-to-rewrite loop compress to 7 days.
- When to use
- Measuring tooling ROI. The metric that proves a stack is operationally fast, not just observational.
- Confused with
- Answer Churn (rate of change, not duration of recovery)
Two teams cannot disagree on a metric they have defined the same way. Most disagreements about citation share are vocabulary collisions, not data problems.
8 confusion pairs the SERP conflates
Each row below is a pair of terms that mainstream articles interchange and that have a bright-line distinction. Print this table, pin it next to your dashboard, and the confusion ends.
| Pair | Bright line |
|---|---|
| Citation Rate vs Citation Share | Rate is absolute (your % of prompts). Share is competitive (your slice of all citations). A brand can grow rate while losing share. |
| Share of Voice vs Share of Model | SoV weights every prompt equally. SoM weights by real prompt frequency. SoM is the commercial truth, SoV is the panel snapshot. |
| Mention vs Citation | Mention = name appears (linked or not). Citation = clickable source link. Mention is awareness, citation is endorsement. |
| Mention Rate vs Mention Frequency | Rate counts how many prompts mention you (breadth). Frequency counts how many times per response (depth). |
| Sentiment Score vs Sentiment Ratio | Score averages polarity (−1 to +1). Ratio counts buckets (% positive). 70% positive at score 0.1 means lots of mild positives, no enthusiasm. |
| Passage Prominence vs Position Prominence | Passage = which block (lead/body/footer). Position = which sentence within. Same citation can be high passage, low position. |
| Engine Spread vs Cross-Engine Coverage | Spread is variance (concentration risk). Coverage is count (presence breadth). One engine at 35% with 5 at zero = high coverage, catastrophic spread. |
| Answer Churn vs Recovery Half-Life | Churn is the rate of change (volatility). Half-Life is the response time (operational speed). Churn is environmental, Half-Life is what your stack does about it. |
Where to go deeper
This glossary is the reference layer of the GEO measurement cluster. The companion articles cover the practice, the tools and the executive reporting layer that all use this vocabulary.
- GEO Tools and Analytics: The Complete Measurement Guide shows the four operating metrics (citation share, engine spread, passage prominence, answer churn) wired together with a 5-minute diagnostic.
- How to Measure GEO Performance: The Weekly Operator’s Playbook wraps these metrics into a 200-prompt set with sampling discipline and a ±2 SD noise floor.
- Best GEO Tools 2026: Honest Teardown of 9 Platforms scores the tools that compute these metrics for you, with the verdict slot per ARR stage.
- Competitor Analysis in AI Search applies Share of Model and Branded vs Unbranded Ratio to the three-competitor baseline.
A glossary is not a vanity exercise. It is the moment your CMO stops asking what the number means and starts asking why it moved. That is the moment the measurement function becomes a decision function.
