Blog
AI visibility metrics explained: mentions, citations, and share of voice
Mike Holp · Published · Updated · Reviewed · 5 min read
AI visibility is not one number. It is a set of observations showing whether answer engines mention your business, recommend it, cite your pages, and place it ahead of alternatives for the prompts your customers ask.
Short answer: Track prompt-level inclusion, recommendation position, sentiment or framing, citations, competitor share of voice, and change over repeated runs. A blended score is useful for a headline, but the underlying answers are what tell you what to fix.
The six AI visibility metrics that matter
1. Mention rate
Mention rate is the percentage of tracked prompts where an engine names your business at all. Track branded and unbranded prompts separately. Branded visibility shows whether the engine recognizes you; unbranded visibility shows whether it connects you to a category, problem, or location.
2. Recommendation rate
A mention is not necessarily a recommendation. Record whether the answer presents you as a good option, merely lists you, or warns the reader away. This distinction matters because a business can appear in an answer without receiving a persuasive reason to choose it.
3. Position and prominence
Position records where you appear in a list or comparison. Prominence adds context: an answer may mention a brand first, explain its strengths, and link to it, or mention it once near the end. Keep the original answer so reviewers can validate the classification.
4. Citation rate
Citation rate measures how often an engine links to or names one of your pages as a source. Citations are different from mentions: an engine can recommend a business without citing its site, or cite a page without recommending the business. Analyze the cited URL, page type, and claim supported by the page.
5. Competitor share of voice
Share of voice compares your appearance frequency with named alternatives across the same prompt set. It is most useful for category and comparison prompts, where the real question is not simply “did we appear?” but “who did the engine choose instead?”
6. Trend and volatility
AI answers can change between runs. Store the engine, prompt, date, answer, citations, and classification so you can compare repeated samples. A one-off result is an observation; a repeated pattern is a better basis for a content decision.
Mentions versus citations
Use this simple interpretation:
| Observation | What it tells you | Possible next action |
|---|---|---|
| Mention without citation | The engine knows the entity but did not show a source | Improve useful, crawlable evidence on the relevant page |
| Citation without recommendation | A page is considered useful, but the answer does not favor you | Clarify positioning, proof, and fit for the prompt |
| Recommendation without your site citation | Your entity has recognition, but first-party evidence is weak | Strengthen the page that should support the recommendation |
| Competitor citation | Another source is supplying the evidence | Study the claim and publish a more complete, accurate answer |
Do not treat any row as a guaranteed ranking recipe. It is a diagnostic pattern that needs review in context.
A practical measurement model
Start with a fixed set of 10–30 prompts covering branded, category, problem, local, and comparison intent. Run the same set across the engines relevant to your audience. For each result, record:
- Engine and date.
- Exact prompt.
- Whether the business appeared.
- Whether it was recommended and its position.
- Sentiment or framing.
- Cited URLs and cited third-party sources.
- Competitors named.
The free VisiScan AI visibility scanner is designed for a baseline. If you need recurring prompt history, compare a monitoring platform after you know which prompts and decisions matter. The best AI visibility tools guide explains that distinction in more detail.
How to turn a metric into a content action
- Low mention rate: check entity clarity, category language, and whether the site explains who you serve.
- Good mention rate but low recommendation rate: add specific proof, outcomes, use cases, and fit criteria.
- Good recommendations but low citation rate: make the supporting page clearer, more complete, and easy to crawl.
- High competitor share of voice: compare the questions and evidence where competitors are repeatedly selected.
- Large run-to-run swings: expand the sample, keep prompts stable, and interpret trends instead of single answers.
FAQ
What is the most important AI visibility metric?
There is no universal winner. For a baseline, mention and recommendation rates show whether you are included and favored. For content work, citation URLs and competitor comparisons usually provide the most actionable evidence.
Is AI visibility the same as a ranking?
No. Many answer engines return generated answers rather than a conventional ordered results page. Position can still be recorded when an answer lists options, but visibility should include mentions, recommendations, citations, and context.
How often should AI visibility be measured?
Run a baseline before making changes, then repeat the same prompt set on a schedule appropriate to the business. Keep the prompt set stable long enough to identify a pattern before adding many new questions.
What should an AI visibility report include?
Include the exact prompts and answers, engine and date, mention and recommendation classifications, citations, competitors, changes since the prior run, and a prioritized action list. A score without evidence is difficult to audit.
Sources
Keep going
Turn the ideas in this article into a measurable baseline for your own site.