Share of Voice in AI

What it measures, and what it does not

Share of Voice in AI counts mentions inside generated answers, not clicks, impressions or ad placements. It shows who an AI assistant actually recommends when a buyer asks it a category question, as opposed to who ranks highest in a search results page. The two can move independently: a brand can lead traditional search rankings and still rarely get mentioned by name in an AI answer, or the reverse. The metric is always relative, your share against specific named competitors for a specific set of prompts, not an absolute score that means anything on its own. It is also a different metric from the share-of-voice figures used in social media or paid media reporting, even though the name is the same; those count posts, impressions or ad spend, not what an AI assistant chooses to say. The denominator matters as much as the count itself: a brand mentioned in eight out of ten answers where only two competitors are tracked tells a very different story than eight out of ten answers where eight competitors are in the running. Two brands can each get named in the same number of answers and still have very different shares, once the size of the competitive set is accounted for.

How it works

The starting point is a set of prompts real buyers would plausibly type, phrased the way a person actually asks, such as "best CRM for a five-person sales team", not just a brand name search. Each prompt runs across the AI assistants your buyers use, repeated over time rather than once, and every brand mentioned in each answer gets logged. Share of voice is the share of those logged mentions that belong to you rather than to a named competitor. Because AI answers can vary between runs, even for an identical prompt sent twice, the metric only becomes meaningful once it is measured repeatedly and averaged, not from a single run. Counting a mention is not always as simple as spotting a brand name; a mention of a specific product line, or of the parent company behind a smaller brand, has to be decided as in-scope or out-of-scope before the numbers mean anything, and that rule needs to stay consistent across every run. Some tracking also records where in the answer a brand appears, since a name mentioned first, with a clear recommendation attached, carries more weight than one listed last among five alternatives, even though both count as a single mention in the simplest version of the metric. Running the same prompt set on a fixed cadence, weekly rather than sporadically, is what turns a handful of individual checks into a trend worth acting on.

Why it matters for AI visibility

Traditional share of voice measured presence in search results or ad auctions, a position that, once earned, tends to hold fairly steady. Share of Voice in AI measures something buyers now act on directly: which brand gets named when someone asks an AI assistant for a recommendation instead of typing a query and comparing links themselves. This is central to AI visibility, because a strong traditional ranking does not guarantee a mention in that generated answer, and the two figures can tell two different stories about the same brand. A buyer who asks an assistant directly for a recommendation has often skipped the comparison step altogether, which means the brand named in that answer starts the conversation with an advantage the runner-up may never get the chance to close. A low overall share of voice is hard to act on by itself; a low share against one specific, named competitor for one specific, recurring prompt gives a team something concrete to address, whether that means closing a content gap or fixing a claim the AI assistant is not picking up on.

Good practices

• Define the prompts that actually matter for your category before measuring anything, based on how buyers really phrase questions.
• Run the same prompt set consistently across every AI assistant your buyers use, not only the one that is easiest to check.
• Track mention rate over time against the competitors you actually compete with in the answer, not a generic industry list.
• Separate simple mention count from percentage share, since a growing prompt set changes raw counts without changing your real position.
• Read sentiment alongside the mention itself; being named is not the same as being recommended.
• Use a tool such as Truffle's competitor tracking to automate the repeated runs a reliable share-of-voice figure requires.

Common mistakes

• Testing a handful of prompts once and treating the result as a stable score, when a single check is only a snapshot.
• Tracking only one AI assistant when buyers actually use several.
• Comparing raw mention counts instead of percentages, which hides how a changing prompt set shifts the picture.
• Ignoring sentiment. A mention inside a negative or lukewarm answer still counts as a mention but is not the same result as a strong recommendation.

Related terms

• Brand Mention: any instance where an AI answer names your brand, the raw event share of voice is built from.
• Citation Rate: the share of answers that cite your site as a source, a related but distinct measure from being merely named.
• Prompt Volume: how many real buyer prompts exist for a category, which sets the size of the base a share of voice is measured against.
• Sentiment in AI Answers: whether a brand mention reads as positive, neutral or negative, a layer share of voice alone does not capture.
• Visibility Score: a combined metric that often folds share of voice together with citation rate and sentiment into one number.

Frequently asked questions