Share of voice, explained

Knowing you were cited in an AI answer is useful. Knowing how much of that answer’s sourcing belonged to you, versus your competitors, is what turns a single citation into a strategy. That is what share of voice measures.

It takes the yes-or-no question of whether you were cited and converts it into a proportion you can compare, track, and deliberately move.

What share of voice actually is

For a given prompt and engine, share of voice (SoV) is your citation count divided by the combined citations of you plus your named competitors. In words, it is your slice of the sources the engine chose to stand behind for that question. If three sources are cited and one of them is you, your share of voice is one third.

The bar above shows the idea. The cited sources for a prompt are a finite pie. Your share is the portion of that pie carrying your name, and the rest belongs to the rivals the engine cited alongside you. The numbers shown are only an illustration of the shape, not a benchmark to aim at.

The most important word in the definition is named. Share of voice is always relative to a competitive set you define. Measure yourself against the wrong rivals, or an incomplete list, and the percentage becomes meaningless. A clean SoV starts with an honest answer to the question of who you are actually competing with for this topic.

Why a proportion beats a yes or no

A simple “were you cited” answer has a blind spot. You can be cited and still be losing badly, because the engine cited four competitors alongside you. You can also lose a citation entirely and not understand whether the category got smaller or a rival took your place. A raw citation count cannot tell those situations apart.

Share of voice fixes this by holding your result against the competitive field at the same moment. It answers the question that actually matters, which is not “did I appear” but “how much of this conversation do I own compared with the people trying to take it.” That reframing is what makes SoV a scoreboard rather than a checkbox. A scoreboard implies a game you can win or lose against named opponents, and that is precisely how AI citation behaves.

How share of voice is measured

Measuring SoV well is a three-step discipline. First, assemble a representative set of prompts that real buyers would ask in your category. Second, run those prompts against each engine and record the cited sources for every answer. Third, for each prompt, count your citations against the citations earned by your defined competitors and express your slice as a proportion.

The quality of the prompt set decides the quality of the number. A handful of vanity prompts that obviously favor you will produce a flattering figure that means nothing. A broad, realistic set that mirrors how buyers actually phrase their questions produces an SoV you can trust and act on. This is why building the right prompts comes before measuring anything.

Measuring share of voice in three steps.

The factors that move your share of voice

• Your own citation rate, every prompt where you move from absent or mentioned to cited adds to your slice.
• Competitor performance, a rival earning new citations shrinks your share even if your own citations are unchanged.
• The size of the cited set, an engine that cites more sources per answer spreads the pie thinner for everyone.
• Model refreshes, when an engine updates, the cited set can be reshuffled, sometimes changing your share overnight.
• The prompt mix you track, adding or removing prompts changes the denominator, so a stable prompt set is essential for honest trends.
• Your competitive set, who you count as a rival directly determines the percentage, which is why the named set has to be deliberate.

When and where share of voice applies

SoV is at its most valuable as a trend and a comparison, not as a single snapshot. A rising share on your priority prompts means you are actively displacing rivals in the cited set. A sudden drop is an early warning that something changed, whether a model refresh, a competitor’s new content, or a shift in how the engine sources that topic.

It also has to be read per engine. Each engine cites a different mix of sources, so your share on one is not your share on another. Winning the conversation on one assistant does not mean you are winning on the next, and a single blended number across engines can hide a serious weakness behind a strong average. Track each engine on its own, and track the same prompts over time, so the movement you see is real signal rather than the noise of a shifting prompt list.

Best practices for tracking it well

1. Define your competitive set deliberately and keep it stable, since it is the denominator of every figure you produce.
2. Build a realistic, category-representative prompt set rather than prompts engineered to flatter you.
3. Keep the prompt set fixed once chosen, so changes in your share reflect the world and not your measurement.
4. Read share of voice per engine, never blended, because each engine cites a different mix of sources.
5. Watch the trend over time, treating a rising line as displacement and a sudden drop as an alarm to investigate.
6. When your share falls, identify which domains gained the slice you lost, so the response targets a real competitor rather than a feeling.

The trap of the one-off check

The honest limit of share of voice is that a single measurement tells you almost nothing. Run your prompts once and you get a number with no context. Is forty-one percent good? You cannot say without knowing whether it was thirty last month or fifty. A one-time SoV is a vanity figure that feels precise and teaches nothing, and it is easy to mistake a lucky snapshot for a durable position.

The way around the trap is consistency. Lock your prompt set and your competitive set, then measure on a steady cadence so each reading is comparable to the last. You cannot manage a number you check once, and you cannot tell whether you are winning or losing a category from a single frame of a moving picture. The value is entirely in the sequence of readings, not in any one of them.

Where to begin and how to keep score continuously

Share of voice is only as good as the prompts behind it, so that is where to start. The free AI Prompt-Pack Generator helps you assemble a realistic, category-representative set of prompts, the kind real buyers actually ask, which is the foundation of any SoV figure you can trust. A strong prompt pack turns share of voice from a guess into a measurement.

From there the work is continuous, because share of voice only means something as a trend. CiteTrack’s citation tracking and share of voice runs your prompts against each engine on a steady cadence, computes your slice against your named competitors per prompt and per engine, and rolls the result into the narrative-control map so you see not only your number but exactly which domains are holding the share you are missing. If you want the underlying mechanics of why engines pick the sources they do, the lesson on how AI decides what to cite is the right next step.