The unit of evidence is a successful answer to a defined prompt under known conditions. A dashboard score without the prompt, answer, platform, date, and denominator is not enough to explain what is happening.
The four parts of AI visibility
Mention
Did the answer name the brand or one of its verified aliases? A mention is the simplest form of visibility, but it does not show whether the brand was recommended or described well.
Position and role
Was the brand the primary recommendation, one option in a list, a passing example, or a comparison target? UnderAI uses metrics such as Visibility and TOP3 to summarize position, while preserving the Answer Snapshot needed to review the role directly.
Sentiment
How did the answer characterize the brand? Public reporting should call this field Sentiment. In UnderAI GEO Workspace, each evaluated Answer Snapshot can carry a Positive, Neutral, or Negative state. That label should be reviewed with the answer text; it is not a substitute for reading material claims.
Citations
Which URLs and domains supported the answer? A citation can reveal an official-site gap, a competitor advantage, or an external publisher that supplies trusted evidence. It does not prove that a particular page section caused the answer.
AI visibility is measured against a question set
There is no useful “visibility across all AI.” Results depend on:
- the prompts tested;
- the platform and product mode;
- the market, language, and date;
- whether the answer was successfully collected;
- how brand aliases and recommendation positions were classified;
- whether prompts were repeated to measure volatility.
A representative Prompt Library should begin with real buyer and stakeholder questions, then be reduced to a stable monitoring panel. Generated prompt ideas can expand discovery, but they should not quietly replace the fixed baseline used to measure change.
Core AI visibility metrics
| Metric | Question answered | Important boundary |
|---|---|---|
| Coverage Rate | For how many active prompts do we have evaluable results? | Collection failure is missing data, not zero visibility |
| Mention Rate | How often is the brand mentioned? | Define the successful-response denominator |
| Visibility | How prominently does the brand appear? | Formula and position rules must stay stable |
| TOP3 | How often is the brand among the first three recommendations? | A list position is not necessarily endorsement |
| Sentiment | How is the brand characterized? | Review the underlying wording |
| Citation Rate | How often do evaluable answers contain citations? | Keep “no citation” separate from unavailable export |
| Official Citation Rate | How often is the target official domain cited? | Canonicalize URLs without deleting response-level facts |
See the complete AI search visibility metrics guide for formulas and missing-data rules.
Example: one prompt, several visibility outcomes
Suppose a team tracks: “What are the best AI visibility tools for a multi-market brand?”
- The brand is absent: this is an evaluable non-mention.
- The brand appears seventh in a list: it has a mention, but not TOP3.
- The brand is recommended first with a positive description: mention, strong position, and positive Sentiment are all present.
- The brand appears but a competitor’s official page is cited: the mention and citation outcomes point to different actions.
- The platform fails to return a usable answer: this is missing measurement and must not enter the denominator as a zero.
This separation is what makes AI visibility actionable.
What can improve AI visibility?
Teams usually need coordinated work across four surfaces:
- Positioning: make the entity, audience, use case, and differentiators explicit.
- Official website: give important questions one complete, accessible page owner.
- External evidence: earn credible third-party coverage where answer engines already source information.
- Measurement: preserve a fixed prompt set and retest after controlled changes.
More content is not automatically the answer. A missing page should be created; an incomplete owner should be strengthened; duplicate pages should be consolidated; and external-source gaps need source work rather than another blog post.
How UnderAI measures AI visibility
UnderAI GEO Workspace organizes Brand Workspaces, market-specific Tracking Projects, competitors, Prompt Groups, and daily platform results. Current international defaults include ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, and Gemini, with enabled platforms configured by project.
The product connects aggregate metrics to Answer Snapshots containing the full answer, mentioned brands, Sentiment, and Citation evidence. UnderAI’s managed AI Visibility Audit then reviews that evidence, identifies gaps, and assigns actions to a page, source, or measurement owner.
Explore UnderAI GEO Workspace or compare AI visibility tools.
What AI visibility looked like in a real category baseline
UnderAI’s July 2026 baseline demonstrated that answer availability, brand presence, and source availability are separate measurements. The AI Search Visibility Metrics page owns the calculations, while the Citation benchmark publishes the full platform results and exclusions.
Calling an active answer environment simply “0% visibility” can hide the source opportunity. A useful visibility review asks whether the platform answered, whether the brand appeared in the right role, and which pages supplied the evidence.
A response-level visibility record
For each successful answer, preserve:
| Layer | Example fields |
|---|---|
| Question | Prompt ID, exact wording, business task |
| Collection | Platform, market, date, run, route, status |
| Brand | Mention, position, role, Sentiment, description accuracy |
| Competition | Mentioned and recommended alternatives |
| Evidence | Canonical Citation URLs and publisher types |
That record makes visibility explainable. A percentage without the underlying answer cannot tell a team what to change.
Research note
The July 2026 figures are from a frozen 16-Prompt US English Panel and should not be used as a cross-industry benchmark. Platform-specific Citation rates are reported in AI Citation Rate Benchmarks. Google’s people-first content guidance provides the quality standard for the public pages intended to improve this baseline.
Frequently asked questions
Is AI visibility the same as SEO visibility?
No. SEO visibility usually summarizes presence in ranked search results. AI visibility evaluates how an entity appears inside generated answers, including mention, recommendation role, Sentiment, and citations.
Does a citation count as a brand mention?
Not automatically. An official URL may be cited without the brand being named prominently, and a brand may be mentioned without any citation. Store both facts.
Can one prompt establish an AI visibility score?
It can provide one observation, not a representative baseline. Use a documented prompt set, successful-response rules, and repeated runs where volatility matters.
Is referral traffic an AI visibility metric?
Referral traffic is a downstream outcome. It is useful, but it does not capture answers that influence a buyer without producing a click.