Which AI Visibility KPIs Should Brands Track: Mentions, Share of Voice or Mention Rate?

What are the main AI visibility KPIs, and how are they calculated?

All of these are calculated over a fixed question set, repeated runs and named engines, so they can be compared over time:

Report each with its base, such as "recommended in 18 of 60 answers (30%)", so readers can see the sample behind the percentage.

What is the difference between AI visibility, share of voice, and citation share?

• AI visibility is the general idea of how present you are in AI answers. Tools often turn it into a single score, combining mentions, positions and citations with their own weights. Useful as a headline if you know the formula, opaque if you do not.
• Share of voice is relative: your mentions as a proportion of all tracked brands' mentions. It shows competitive position, but it depends on which competitors you track. It can rise because a competitor disappeared, not because you improved.
• Citation share is about sources, not brands: how many of the links in answers point to your domain, compared with all links. You can have high share of voice and low citation share when engines recommend you based on third-party reviews rather than your own pages.

The three answer different questions. Visibility asks "are we there?", share of voice asks "are we there more than rivals?", and citation share asks "are our pages the evidence?" How to calculate share of voice properly is covered in what AI search share of voice is and how to measure it.

What KPIs should replace keyword rankings for measuring AI search?

Rankings measure position in a stable list. AI answers are generated fresh and vary between runs, so position alone misleads. The replacements are:

1. Recommendation rate on buyer questions, instead of average ranking position.
2. Question coverage, instead of the number of keywords in the top ten.
3. Citation rate, instead of organic clicks as the measure of your pages being used.
4. Share of voice against named competitors, instead of competitor ranking comparisons.
5. Accuracy rate, which rankings never measured and which matters because AI answers can describe you wrongly.

Keep rankings for Google's classic results. They still explain most search traffic, and they feed the pages AI systems retrieve.

What AI visibility KPI should I report to a CMO each quarter?

One headline and one outcome, with context:

• Headline: recommendation share on priority buyer questions against the main competitor. For example: "Across 40 priority questions in ChatGPT, Google AI Mode and Perplexity, we were recommended in 31% of answers, up from 14% last quarter. Competitor A was at 44%."
• Outcome: AI-sourced pipeline or revenue, from referral data and self-reported attribution. For example: "AI assistants were cited as a source in 11% of new enterprise leads."
• Context: one line on accuracy ("no incorrect pricing statements found this quarter") and one on the biggest change and its cause.

That gives leadership a number they can track, a competitor benchmark and a link to revenue, without a dashboard of twenty metrics. The rest belongs in the appendix for the marketing team.

How do you build a KPI hierarchy?

It helps to see the KPIs as layers, from what engines do to what the business gains:

Problems usually flow downwards. Access problems cap presence, and weak presence caps preference. So when an outcome KPI falls, check the layers above it first.

What traps come with each KPI?

• Raw mention counts grow when you add questions or runs; always use rates.
• Brand-name questions inflate mention rate, because asking about your brand guarantees a mention. Keep them in a separate accuracy set, as explained in whether brand-name prompts inflate AI visibility scores.
• Share of voice shifts when you change the competitor list; keep it fixed for a period.
• Citation share can be boosted by irrelevant pages; count only citations on questions that matter.
• Composite scores hide which component moved; report the parts too.
• Single runs are noisy; use at least three runs per question per engine.
• Averages across engines can hide a collapse in one; report by engine.

How many questions and runs do KPIs need?

Enough that a change of a few answers does not swing the headline. As a starting point, 30–60 buyer questions per market, three runs each, in the two to four engines your customers use. At that size, treat quarter-on-quarter changes of less than about 15 percentage points in the pooled headline (about 20 points for a single engine) as noise until a rerun confirms them. Changes of two or three points are usually noise. Report ranges or confidence intervals when samples are small. How AI visibility in Google specifically can be measured is covered in how to measure brand visibility in Google AI Overviews.

What do people ask about AI visibility KPIs?

Is mention rate or share of voice more important?

Mention rate shows your own progress; share of voice shows competitive position. Track both, and lead with recommendation rate for business audiences.

Should sentiment be a KPI?

Track it as part of accuracy and positioning. A mention that warns against you counts in mention rate but hurts the business.

Can Search Console provide AI visibility KPIs?

It provides impressions in AI Overviews and AI Mode by page, which is useful exposure data. It does not show mentions of your brand or competitors.

How often should KPIs be reported?

Monthly for the marketing team, quarterly for leadership, using the same questions and engines each time.