Four figures can describe four different problems. A brand can appear in many tracked answers and still trail competitors. It can also land late when mentioned and rarely have its own pages cited.
One score would hide the decision inside that pattern.
Key takeaways
- Zumi's AI Visibility model contains four separate signals: mention rate, share of voice, average position, and citation share.
- Mention rate measures presence, while average position measures prominence among brands that appear.
- Share of voice compares a brand with a declared competitor set. Citation share measures how often the brand's own pages support the answer.
- Every result needs a measurement contract covering the prompt set, engine, geography, date, sampling, and denominator.
- AI Visibility describes observed discovery. It does not prove traffic, conversion, pipeline, or revenue.
Why can't one visibility score do the job?
A composite score combines several observations through a weighting rule. That can make a dashboard easier to scan, but it also makes the result harder to audit. A change in the headline number does not reveal whether the brand appeared more often, moved higher, gained ground on competitors, or earned more citations.
The four-signal model keeps those movements visible. Each signal answers a different operating question, and none can stand in for the other three.
Zumi is an AI Search Intelligence Platform. Its AI Visibility model reports mention rate, share of voice, average position, and citation share as separate descriptive signals, refreshed daily. The separation is part of the method, not a presentation choice.
What does each signal measure?
The signals describe presence, competitive standing, prominence, and source support. Each needs a declared set of eligible answers before it can be interpreted.
- Mention rate: How often a brand is mentioned in tracked answers. It measures presence, but one mention does not establish stable coverage.
- Share of voice: How often a brand appears relative to tracked competitors. It measures competitive presence, but it cannot be compared across different competitor sets or denominators.
- Average position: Where a brand lands among the brands mentioned in an answer. It measures prominence, but it is not a stable search ranking.
- Citation share: How often a brand's own pages are cited as sources. It measures owned-source support, not every third-party page that mentions the brand.
Mention rate measures coverage
Mention rate starts with an eligible answer set. The numerator is the set of answers that mention the brand, and the denominator is the full set of eligible tracked answers in scope.
The signal does not describe where the mention appears or what competitors appear beside it. A high rate can coexist with weak prominence and low citation share. A low rate can also hide a strong position within the few answers where the brand appears.
Share of voice measures relative presence
Share of voice places the brand inside a competitive field. The figure depends on the tracked brands, prompt panel, counting rule, and engine scope. A change to any of those inputs changes the comparison.
Tools use more than one share-of-voice denominator. Some divide the brand's mentions by all tracked brand mentions. Others use all eligible answers.
The label is incomplete unless the denominator is disclosed.
Average position measures prominence
Average position records where the brand appears among other mentioned brands. A lower numeric position usually means earlier placement, but the counting rule must explain how prose, unordered lists, and absent brands are handled.
This signal is not a search rank. AI-generated answers can change between repeated requests, and a prose answer does not always produce a clean ordered list. Position therefore belongs beside mention rate and the underlying answer evidence.
Citation share measures owned-source support
Citation share describes how often the brand's own pages are cited among the sources in scope. It does not count every source that happens to discuss the brand as an owned citation.
The source distinction changes the action. Low owned citation share can point to a missing or unusable page. Strong third-party activity can point to an earned-media opportunity.
The cited URLs provide the diagnosis that a percentage cannot.
Which scope must travel with every number?
An AI visibility result is incomplete without its measurement contract. The contract identifies the object that was observed and makes a later comparison possible.
- Prompt panel: Exact questions, stable IDs, intent labels, and version.
- Engine or answer product: The named interface or product measured, kept separate where behavior differs.
- Market: Geography and language.
- Test state: Date range, account state, personalization state, and fresh-session rule.
- Sampling: Repetitions, cadence, failed answers, and exclusions.
- Entity set: Canonical brand names and declared competitors.
- Counting rule: Mention, position, citation, and denominator definitions.
A figure from ChatGPT cannot be silently blended with a figure from Google AI Overviews. A global prompt panel cannot be compared with an India-specific panel as though the market stayed fixed. A new competitor set also starts a new share-of-voice series.
The contract does not remove variation. It makes the variation visible and prevents a changed ruler from being reported as progress.
How should the four signals be read together?
The signals work as a diagnostic sequence:
- Mention rate establishes presence. A weak rate identifies questions and engines where the brand is absent.
- Share of voice adds the competitive field. A brand can appear often and still trail a competitor across the same panel.
- Average position tests prominence. A mentioned brand can remain peripheral inside the answer.
- Citation share traces owned support. The source record shows whether the brand's own pages are helping form the answer.
The combinations matter more than an isolated movement. Rising mention rate with falling average position can mean broader but weaker inclusion. Stable mention rate with rising citation share can mean the source layer changed before broader presence did.
No combination proves why the movement occurred. The raw answers, cited pages, shipped work, and unchanged measurement contract support an interpretation. They do not turn an observed comparison into a controlled causal test.
Which actions can follow the measurement?
Each signal opens a different investigation.
- Low mention rate: Content, product marketing, or public relations examines missing category coverage, unclear entity information, and absent third-party support.
- Adequate mention rate but weak share of voice: Strategy or competitive intelligence reviews rival presence by engine, question, and source.
- Frequent mentions but weak average position: Content and product marketing reviews the selection criteria and sources behind earlier recommendations.
- Low owned citation share: Web, content, or technical search reviews missing pages, inaccessible evidence, and stronger third-party sources.
- Strong third-party citation activity with weak brand presence: Public relations or partnerships reviews gaps in the sources the engine already uses.
Two related pieces go further: the dark funnel in AI search, measured covers the buyer research these signals capture before any click, and AI visibility is not the product covers why a number alone doesn't keep a program funded.
Zumi's Recommendations feature groups prescribed actions into Owned Media and Earned Media. The platform identifies what content to create and where coverage may be needed. The responsible brand or agency team performs the writing, publishing, and outreach.
What can't these signals prove?
The four signals describe what appeared in observed AI-generated answers. They do not measure visits, conversions, pipeline, revenue, or incremental commercial impact.
Those outcomes belong to separate analytics and attribution systems. Referral data, self-reported discovery, customer-relationship records, and controlled commercial analysis may add context, but they should not be folded into the AI Visibility definition.
The practical standard is narrow. The four signals stay separate, and the measurement contract stays attached. Every action remains traceable to the answers and sources that prompted it.
The manual baseline playbook shows how to create that first measurement without buying a platform.