How to read a share-of-voice chart (and why 30% doesn't mean what you think)

Picture a fairly typical share-of-voice chart. Five brands in a category, ranked by how often they show up across a set of tracked AI prompts: the leader sits at 31%, the next two brands trail at 24% and 18%, and the smaller players round out the bottom at 15% and 12%. It looks clean, complete, and easy to act on. Someone in a meeting will point at the 31% and say "we're winning." Someone else will point at their own 12% and ask what's wrong.

Both reactions skip a question the chart itself never answers: 30-something percent of what, measured how, against whom, and is that even a number worth being happy or upset about in the first place? Share of voice is one of the most useful metrics in AI visibility tracking, and also one of the easiest to misread, because the chart shows you a single clean bar and hides every judgment call that produced it.

What a share-of-voice chart is actually measuring

In AI visibility terms, share of voice is usually some version of: out of every time a brand in your category could plausibly have been mentioned across a set of tracked prompts, what percentage of those instances were you, versus a competitor, versus nobody at all. That definition has at least four moving parts, and every one of them is a choice a person or a platform made before the chart was ever drawn: which prompts counted, which competitors counted, over what time window, and what exactly counts as an "instance," a mention, a citation, or both blended into one number. Change any one of those four and the same underlying reality can produce a very different-looking bar chart.

The 40-year-old metric hiding underneath a new dashboard

Share of voice did not originate in AI search. It's a core concept in advertising economics, and the most important thing that decades of research on it established has almost nothing to do with the raw percentage itself. Researchers at the UK's Institute of Practitioners in Advertising, most notably Les Binet and Peter Field working with the IPA's effectiveness databank, found that a brand's share of voice only predicts growth relative to its existing share of market, not on its own.¹ The gap between the two, called excess share of voice, is what actually correlates with whether a brand grows or shrinks: their analysis found that for every 10 percentage points a brand's share of voice exceeds its share of market, it tends to gain roughly half a percentage point of market share per year, compounding over time.¹ The relationship runs both ways. Brands whose share of voice fell below their share of market lost market share in roughly 80% of the cases studied.¹ LinkedIn's B2B Institute, working with Binet and Field directly, later confirmed the same pattern holds in B2B categories, if anything slightly more strongly than in consumer markets.²

Translate that lesson directly into AI visibility, and the 31% at the top of our example chart stops being a single, self-contained fact. If that brand only holds 10% of the actual market (real customers, real revenue), a 31% share of voice is a strong positive signal, arguably a leading indicator that its market share is about to grow. If that same brand already holds 45% of the market, a 31% share of voice is a warning sign: it's underrepresented in AI answers relative to how big it actually is, and on the historical pattern, that's a brand more likely to be losing ground than gaining it.

None of this means the underlying math transfers perfectly from TV advertising spend to AI citation share. It hasn't been tested at that level of rigor yet, since AI search itself is only a few years old. But the core discipline, always reading a share-of-voice number against your actual size in the market rather than in isolation, is exactly the habit missing from most AI visibility reporting today.

Reason 1: the denominator changes everything

Share of voice is a fraction. Most of the disagreement over what a number "really means" comes down to disagreement over what's sitting in the denominator, and dashboards rarely show their work here.

The same brand, tracked against the exact same set of AI responses, can produce three meaningfully different percentages depending on which of these three a dashboard defaults to.

Put real numbers on it. Say you ran 100 prompts in a category. Your brand was named in 20 of them. A competitor was named in 25. The remaining 55 responses named no brand at all, just generic advice. Measured as share of all tracked prompts, you're sitting at a modest 20%. Measured as share of mentions, counting only the 45 responses where any brand showed up, your number jumps to roughly 44%, more than double, without a single thing about your actual visibility changing. Both numbers are technically correct. Only one of them is being shown to you, and a dashboard rarely volunteers which.

Reason 2: the competitor set is an editorial choice

Nobody hands you a definitive list of "your competitors." Someone, a platform's default settings or your own team, decided which brands belong in the comparison, and that decision changes the math directly, since share of voice is always relative to whoever else is in the chart. Include only your three closest, most obvious rivals and your own share looks larger simply because there's less competition to divide the pie with. Include a dozen adjacent players, including some only tangentially in your category, and the same brand's share compresses, even though nothing about its actual visibility changed. A share-of-voice chart is never wrong, exactly, but it's always answering the question "relative to this specific list," and that list is worth checking before trusting the number.

Reason 3: a snapshot hides enormous run-to-run noise

Even with the denominator and competitor set held perfectly constant, there's a third problem: AI models don't give the same answer every time you ask them the same question. Independent testing of this directly has found that running an identical brand-recommendation prompt through ChatGPT 100 times can return the exact same list of cited brands in fewer than 1 out of 100 responses.⁴ Separate analysis of AI Overview responses found the cited sources changed entirely on 70% of exactly repeated, identical queries, with an average of 45.5% of citations swapped out for different ones between one response and the next.⁵ A share-of-voice chart built from a single day's sampling, or worse, a single run per prompt, isn't measuring your visibility. It's measuring one noisy draw from a genuinely random process, and treating it as a stable fact is a bigger error than almost any of the definitional issues above.

Reason 4: category size and category value aren't the same thing

Owning 30% of a conversation that barely happens is worth less than owning 10% of one that happens constantly and drives real buying decisions. A share-of-voice chart typically blends every tracked prompt together into one number, which quietly treats a rarely-asked, low-intent question and a frequently-asked, high-intent buying question as equally important. In practice they're not. A brand can post an impressive-looking share-of-voice number built mostly from long-tail, low-value prompts, while a smaller-looking competitor holds a thinner but far more valuable slice concentrated in the handful of prompts that actually precede a purchase. The chart alone can't tell you which situation you're in. You have to look at what's actually inside that percentage, not just its size.

Here's a concrete version of that trap. Imagine a project-management software brand tracking 40 prompts. Thirty of them are broad, informational questions like "what is project management software," asked constantly but rarely tied to an active buying decision. The other ten are sharp, comparative questions like "best project management tool for a 20-person agency switching from spreadsheets," asked far less often but almost always by someone close to signing up for something. A brand that dominates the 30 broad prompts and barely registers on the 10 comparative ones can still post a headline share-of-voice number in the 60 to 70% range, built almost entirely on volume rather than value. A smaller competitor holding just 15% overall, but concentrated entirely in those ten comparative prompts, is arguably in the stronger commercial position, and a single blended chart would never show you that.

Why this matters beyond the chart itself

None of this is only an academic distinction. Teams make real resourcing decisions off these numbers: whether to invest more in content for a category, whether to treat a competitor as a genuine threat, whether to report a quarter as a win or a loss internally. A misread share-of-voice chart doesn't just produce a wrong number on a slide, it can send a real budget in the wrong direction, either overinvesting in a category where you're already dominant relative to your actual market position, or underreacting to a competitor who looks small on a blended chart but is quietly winning every prompt that actually matters.

A worked example: two charts, same data

Here's the same brand's actual mention count, shown two different ways, to make the competitor-set problem concrete rather than abstract.

Both charts use the exact same raw mention count for "your brand." Nothing about your actual visibility changed between them. Only the list of brands being divided against did.

How to actually read a share-of-voice chart

• Check the denominator definition

  Is this share of mentions, share of citations, or share of all tracked prompts? Ask, don't assume.
• Check who's in the competitor set, and who isn't

  A short list inflates everyone's number. A long list compresses it. Neither is wrong, but both need to be stated.
• Check the sample depth and date range

  A single day's snapshot, or a handful of prompt runs, carries far more noise than a percentage implies. Look for a chart built on repeated sampling over time, not one run.
• Compare it against your actual market share, not in isolation

  A given percentage is a positive or a warning sign almost entirely based on how it compares to how big you actually are.
• Look at the trend, not the snapshot

  A single bar chart is a photograph. What matters is whether that number is climbing, flat, or sliding over the past several weeks.
• Segment by prompt value, not just topic

  A blended, category-wide share of voice can hide a much stronger or weaker position on the specific high-intent prompts that actually drive revenue.

None of this means share of voice is a bad metric. It's one of the more genuinely useful numbers in AI visibility tracking, precisely because it forces a comparison against competitors rather than looking at your own visibility in a vacuum. The point isn't to distrust the chart. It's to ask the four or five questions above before deciding what the number is actually telling you, the same discipline advertising researchers had to learn the hard way, decades before anyone was measuring share of voice inside a chatbot's answer.

The next time someone puts a share-of-voice chart in front of you, the single most useful question to ask out loud isn't "is this good." It's "relative to what." That one question forces the denominator, the competitor set, the sample size and your actual market position all back onto the table, which is exactly where a number this important belongs.