Reference

Share of voice inside AI answers

We compete with Scrunch. We are putting them at the top because the denominator is the whole argument, and theirs is better than ours.

What is share of voice inside an AI answer, and what is it not?

The formula is unremarkable and identical to the one marketing has used for decades: your brand’s mentions divided by total mentions in the measured set, times one hundred. Everything interesting is inside “the measured set”.

In classic search that set was bounded. There were keywords, they had volumes, and you could enumerate them. In AI answers there is no such list. Dan Taylor put it plainly: the universe of possible AI prompts is effectively infinite, and the tools reporting share of voice are presenting a closed sandbox as if it were the open web.

So an AI share of voice figure is not a market share. It is your share of a sample that somebody chose. The only honest version of the metric is one where you can see who chose it and how large it was.

Why do two tools report different share of voice for the same brand?

The denominator problem

Most tools build the sample from prompts you supply. That means your share of voice is measured inside a question set written by your own marketing team, which will over-represent the questions you already win and under-represent the ones you have never thought to ask.

Scrunch’s answer is to ship a library alongside yours. Starter is $250 a month on annual billing for 350 custom prompts plus 1,000 industry prompts and 3 personas. Growth is $417 a month annual for 700 custom plus 2,500 industry prompts. The industry set is the part that matters: it is a denominator you did not author.

Ahrefs takes a different route to the same problem, building Brand Radar on observed monthly prompts rather than synthetic ones, at $398 a month for select platforms and $699 for all. Profound’s Conversation Explorer is a third route, surfacing what people actually ask in a category rather than what you guessed.

The competitor list you wrote yourself

The second failure is quieter. If your tool asks you to name your competitors at setup, your share of voice is computed inside a pool you defined. Brands that beat you but never occurred to you are simply absent from the denominator, and your percentage is flattering by construction.

Scrunch is also the only vendor in this set that runs a public comparison microsite rating pairs of tools it is not part of, on the stated grounds of wanting to remove itself from the bias. Whatever you think of the tactic, it is the correct instinct and the rest of the category, us included, should match it.

The remaining two reasons are run count and platform mix. On run count, SparkToro’s 2,961-run study found under a 1 in 100 chance that two responses to the same prompt name the same set of brands, which makes a single-run share of voice a coin toss dressed as a percentage. On platform mix, Semrush’s index of 126 million prompts found only 36 of 1,200 brands held a top-100 position on every platform in every month.

Which tools measure AI share of voice best?

Ranked on the quality of the denominator and the transparency of the sample. Prices from vendor pages, 2026-08-20.

If your budget is zero, HubSpot’s AI Search Grader scores share of voice as 10 points of a free 100-point check. It is a snapshot, not a tracker, but it costs nothing and it will tell you whether you have a problem worth paying to measure.

What should you measure instead of, or alongside, share of voice?

Taylor proposes three replacements, and they are better metrics than the one they replace. Share of mentions, how often you appear at all in the category’s answers. Share of recommendations, how often you appear when the buyer explicitly asks for a shortlist, which is the only cut with purchase intent in it. Share of narrative, what the answer says about you, because being named often as the budget option is not the same win as being named rarely as the safe one.

The third is the one most teams have no instrumentation for and most need. There is also good reason to weight the second heavily: review platforms take a larger share of citations as buyer intent rises, and G2 with Capterra, Software Advice and GetApp together hold about 84% of citations in that category.

Two constraints to keep the exercise in proportion. AI referrals are roughly 1% of total site visits, though they convert near 7%, and only about 1% of users click a link inside a Google AI summary. Share of voice measures whether you are in the consideration set. It does not measure traffic, and it should not be reported next to traffic as though it predicted it.

Where does Rovoki fit, and where does it not?

Rovoki reports share of voice across ChatGPT, Perplexity, Gemini and Claude, per engine and never averaged without the split beside it, in 18 countries and 31 languages or worldwide, and publishes the raw answer behind every question it counted so the percentage can be opened rather than trusted.

Three things it does not do. It samples n=1 per run, so a Rovoki share of voice figure is directional and we will not describe it otherwise. It does not ship confidence bands, which is what we are building toward and is not live today. It has no persona model and no industry prompt library, which means your denominator is the one you supplied. That last point is exactly the failure this page is about, and it is why Scrunch is at the top of the table rather than us.

If the question you are trying to answer is “what is my share of the conversations I do not know about”, buy Scrunch. If the question is “can I see the evidence behind every number on this dashboard”, that is the one we built for.

There is a broader reason to care about being named rather than linked. Across 75,000 brands, unlinked branded mentions correlated with AI visibility at 0.664 against 0.266 for domain rating, and the authors are explicit that correlation is not causation. Share of voice, done honestly, is a measure of how often other people say your name. That has always been the thing worth counting.