CapstonAI · AI Visibility
Mention rate, citation rate, share of voice and sentiment — the four metrics that define AI visibility, with clear definitions, formulas and benchmarks.
Many tools report one number. It’s convenient and almost useless for action: a score that ticks down doesn’t tell you whether you lost a citation, fell behind a competitor, or got described inaccurately. The four metrics below decompose AI visibility into things you can actually fix. This page is the measurement layer of the broader AI visibility pillar.
The deeper problem with a single score is that it collapses independent outcomes that move for different reasons. Mention rate is driven by how strongly the engines associate your brand with a category; citation rate by how quotable and parseable your pages are; share of voice by what competitors do; accuracy by how clearly your public information is structured. Roll those into one figure and you lose the diagnosis: you can see that something moved, but not what, or what to do about it. Keeping the metrics separate is what turns measurement into a work plan instead of a vanity dashboard.
| Metric | What it measures | How to calculate |
|---|---|---|
| Mention rate | How often you’re named in answers | prompts where you're mentioned ÷ total prompts |
| Citation rate | How often your site is linked as a source | prompts where you're cited ÷ total prompts |
| Share of voice | Your presence vs competitors | your appearances ÷ all brand appearances |
| Sentiment & accuracy | What’s said about you, and whether it’s correct | qualitative scoring + factual check |
The baseline: are you in the conversation at all? A low mention rate means the engines don’t associate your brand with your category — usually an entity or authority problem.
Stronger than a mention, because a citation can drive referral traffic and signals the engine trusts your page as a source. Citations reward quotable, well-structured, sourced content.
The competitive metric. Visibility is relative — appearing in 40% of answers means little until you know a rival appears in 80%. Always benchmark 1:1 against named competitors, not an industry average.
The qualitative layer. A confident but wrong answer about your pricing or features is a visibility problem of its own — you’re seen, but misrepresented.
Say you run a fixed set of 20 category prompts across one engine. Your brand is named in 8 of the answers, linked as a source in 3, and across all 20 answers there are 50 total brand appearances of which 8 are yours. That gives:
The gap between a 40% mention rate and a 15% citation rate is the story: the engine knows your brand but rarely trusts your page enough to link it. That points the fix at quotable, well-sourced content — not at brand awareness. The 16% share of voice then sets the competitive ceiling: if a rival sits at 40%, you know exactly how much ground there is to take.
| Weak metric | Likely cause | Where to act |
|---|---|---|
| Low mention rate | Engines don’t associate you with the category | Entity signals, authority, category content |
| Mentions high, citations low | Content isn’t quotable or parseable | Answer-first passages, schema, sourcing |
| Low share of voice | Competitors win the same prompts | Beat the specific cited sources, prompt by prompt |
| Poor sentiment/accuracy | Outdated or unclear public info | Correct on-site facts, structured data, about pages |
Define a representative prompt set, run it across ChatGPT, Perplexity, Gemini and Google AI Overviews, and record each metric for your brand and competitors. Repeat on a schedule — single runs are noisy because answers vary. For the continuous version, see the AI visibility tracker; to act on weak metrics, the AI visibility tool connects measurement to fixes.
A single blended number across engines hides the decisions that matter. ChatGPT, Perplexity, Gemini and Google AI Overviews each retrieve and cite differently, so the same brand routinely scores well on one and poorly on another. Calculate every metric per engine, then read the spread:
Segmenting also stops a strong engine from masking a collapsing one in your reporting — the most common way a blended score misleads. In practice this is why monitoring brand mentions in ChatGPT separately from the other engines is worth the extra column in your report.
There’s no universal threshold — it depends on category competitiveness and how many brands the engines typically name. The honest benchmark is directional: each metric should be rising over time and gaining on named competitors. Judge yourself against your own trend and your real category rivals, not an abstract industry average.
Mention rate, citation rate, share of voice, and sentiment/accuracy. Together they describe how AI engines surface your brand — far more usefully than a single combined score.
Mention rate is how often you’re named in answers. Citation rate is how often your site is linked as a source. Citations are stronger because they can drive traffic and signal trust.
Divide your brand’s appearances by the total brand appearances across your prompt set. It’s the relative, competitive view of your presence.
There’s no universal threshold — it depends on your category. The reliable benchmark is directional: metrics rising over time and gaining on named competitors.
A single score hides what changed. The four metrics tell you whether you lost a citation, fell behind a competitor, or were described inaccurately — each with a different fix.
On a recurring schedule, because AI answers vary between runs. Trends across multiple runs are more reliable than a single snapshot.
Run a free scan to see your mention rate, citation rate and share of voice.
Put these metrics to work with real-time SEO KPI tracking and a dedicated AI visibility tool.
