Mentions, Citations, Share of Voice: You've Been Measuring the Wrong Thing

Your AI-visibility dashboard tracks three things: mentions, citations, share of voice. All three of these AI visibility metrics answer one question: does the machine know you exist? None of them answers the question that decides revenue: when a buyer asks who is best, does the machine choose you?

The short version: Mentions, citations and share of voice measure existence, not preference. AI citation lists churn: roughly 30% of brands named in one answer survive into the next identical answer, and 20% survive five runs. The stable layer underneath is distinctiveness: whether the machine can tell you apart and attribute you by name. Measure that cause, engineer it, and the mentions follow. Chase the mentions directly and you fund the exact work that erases you.

Existence metrics in a choosing market

Analytics has been here before. For a decade the web measured pageviews, until it became obvious that pageviews told you people arrived and nothing about whether the product worked. The metric was real. It was measuring the wrong layer.

AI visibility is repeating that history. Mentions, citations and share of voice count appearances in an answer set. They are the pageviews of the answer era: true numbers about the surface, silent about the mechanism underneath.

And the surface is noisy. Stability studies of AI answers in 2026 found that only about 30% of brands cited in one response appear again in the very next response to the same query, and around 20% persist across five runs. A weekly mentions report is substantially a report on sampling noise. Teams celebrate a spike, panic at a dip, and neither was a signal.

What the machine holds stable

Underneath the churn, answer engines are consistent about one thing: they name what they can distinguish and trust. The concentration proves it: 3% of brands capture 71% of AI recommendations, and in most categories the citations collapse into three to five names. Those few are not the brands with the most mentions last week. They are the brands the machine can attribute: content recognisably theirs, proof only they can make, an identity that resolves by name.

Being retrieved is not being chosen. Retrieval is a lottery ticket; attribution is a position. The dashboards measure the tickets.

The wrong metric buys the wrong work

Metrics direct budgets, and existence metrics direct budgets to volume: more pages, more posts, more AI-assisted production. That work backfires twice.

First, AI-default production converges toward the category average, so every additional piece makes you harder to tell apart. AI can average your voice; it cannot evolve it. Second, your own domain was never the main battlefield: roughly 85% of what AI assistants say about brands comes from third-party sources. We audited a brand with 31,725 indexed pages and near-zero presence on the community surfaces engines trust most. Its mention chart looked respectable. Its position was invisible.

So the symptom metric funds the disease: more volume, more sameness, less distinguishability, fewer names the machine can hold onto.

Measure the cause instead

Symptom metric (existence)	Cause metric (choice)
Mentions per 100 answers	Distance from your category's average voice
Citation count	Blind attribution: logo off, still recognisably you?
Share of voice	Proof density only you can claim
Answer position	Footprint on the third-party surfaces engines trust

Four measurements replace the dashboard:

1. Distance from the category average. How far your language sits from the centre every competitor's AI is writing toward. This is Voice, measured, not described.
2. Attributability. Strip the identity marks and test whether a machine still assigns the content to you. This is your Entity working or failing.
3. Proof only you own. Named methods, your data, your experts, your decisions. Topic Authority is what the machine trusts enough to repeat.
4. Third-party footprint. Presence where the engines read: communities, reviews, press. The away game, counted.

The Brand Distinctiveness Index packages these into one 0-100 score against your real competitors, every component backed by cited evidence from your own content. That is the discipline we built at Ivanooo: measure the cause, engineer it, then watch it daily, because judgment drifts back to the machine default without noticing.

The question for your next reporting meeting

If our mentions doubled next quarter, would the machine be any more able to tell us apart? If the answer is no, the dashboard is up and the position is unchanged. Find out what the cause metrics say about you: paste your URL, get your weaknesses ranked, with the evidence. No call required.

FAQ

Are AI mentions worthless? No. Mentions are the outcome to verify, not the lever to pull. Track them quarterly as proof the cause work is landing; manage the distinctiveness that produces them.

Why do AI citations change between identical queries? Answer engines sample: only about 30% of cited brands persist into the next identical answer, and 20% across five runs. Week-to-week mention movement is mostly noise, which is why it makes a poor management metric.

What should replace share of voice? Four cause metrics: distance from the category average, blind attributability, proof density only you can claim, and third-party footprint. The Brand Distinctiveness Index combines them into one 0-100 score.

What is the Brand Distinctiveness Index? A 0-100 measure of how distinguishable your brand is to the machines that decide recommendations, scored against your real competitors, with cited evidence behind every component.

Does this replace AEO? It sits underneath it. AEO plumbing (schema, structure, crawlability) gets you retrieved. Distinctiveness gets you chosen. Retrieved is not chosen.