Best tool for measuring share of voice in AI answers, compared on whether the vendor publishes a denominator

Share of voice is a fraction, and in this market the numerator is easy and the denominator is where the money goes. So the useful way to shop is to ask each vendor one question: what count will my percentage be divided by, and is that count published. Read on 31 October 2026, five vendors publish a number you can divide with, and one publishes a sample size, which is a different and better thing.

The short recommendation. Buy Evertune Pro if a board will see the percentage, because it is the only vendor on this roster that publishes how many times it asks each prompt. Buy AthenaHQ Starter or Scrunch Core if you want a clean, countable denominator at a published price. Use Ahrefs Brand Radar when the share you want is market level rather than brand level, because it divides against an index somebody else already ran.

What each vendor publishes as a count, read 31 October 2026

Vendor and plan	The count it publishes	What your share would be out of
Evertune Pro, 800 dollars a month billed annually	100,000 prompt credits a month, 100 samples a prompt from this tier	readings, with a stated sample size per prompt
Evertune Starter, 60 dollars a month	5,000 prompt credits a month, sampling row not included	readings, one look per prompt
AthenaHQ Starter, 295 dollars a month	3,600 credits, one credit is one AI response, unlimited competitors	responses
Scrunch Core, 250 dollars a month	125 prompts, 5,000 responses a month, 5 competitors	responses, against a capped competitor set
Goodie Core, 295 dollars a month	110 prompts on 5 engines, collected daily	responses, computed from cadence rather than stated
Ahrefs Brand Radar, index from 199 dollars a month	519 million or more prompts in the index, plan quotas of 150 to 600 checks	a market index, not your own question set
Conductor, second plan upward	2,500 AI search credits a year, credit not defined on the page	not computable from the page
Profound	50 prompts daily for 7 days on the trial, Enterprise custom	not computable from the page
AI Knows Us, ours	100, 200 or 300 buyer questions watched, 1,200 to 3,500 scheduled answers a month	answers, on five named apps

Why the sample count outranks the credit count

Two vendors can both report that you hold 12 per cent of answers in your category and mean completely different things, because an assistant does not give the same answer twice. If each prompt was asked once, your 12 per cent is one roll of the dice per question. If each prompt was asked a hundred times, it is a distribution.

Evertune is the vendor that publishes this, and it also publishes where it starts: the row reading statistical significance from day one, 100 samples per prompt, carries a dash against Starter and a tick against Pro and Enterprise. Its home page states a margin of error of plus or minus 2 points against typical competitors at plus or minus 31, which is the vendor's own claim about unnamed rivals and is reported here as read rather than independently checked. Even treated as marketing, the shape of the claim is the right shape, and nobody else on the roster prints a sample count at all.

The denominator question has three defensible answers and the choice changes the number by an order of magnitude, which is worked through in three legitimate denominators and why ours reads 0.5 per cent. Take the vocabulary from there into the demo, because a vendor that cannot say which of the three it uses is reporting a number nobody can defend in a meeting.

The competitor cap decides whether a share means anything

A share of voice needs a field, and two vendors publish the size of the field. Scrunch Core allows 5 competitors, Conductor publishes 5, 25 and 75 or more across its three plans, and AthenaHQ publishes unlimited competitors on its 295 dollar Starter plan. In a category with nine real suppliers, a five competitor cap gives you a share of a field you picked rather than a share of the market, and that limitation sits inside the cheapest plans of the two vendors most likely to be on your shortlist.

Ours does not publish a competitor cap, which is a gap on this criterion and is worth saying rather than skipping. What our plan table does publish is who gets named instead of you and which page won them the slot, from the Growth plan upward, so the competitor set is derived from the answers rather than configured in advance.

Where rivals beat us on this specific job

Three lines, plainly. Evertune publishes a sample count and we do not publish ours on the pricing page. AthenaHQ publishes an unlimited competitor set at a published price where we publish no cap at all. And Goodie and Evertune both collect daily, where we read your priority questions weekly and the rest at 30 and 90 days, so a daily share is something they can produce and we do not sell.

Where we are useful on this job is the denominator itself. We publish the three denominators, the reading count behind our own figures, and the per engine split that makes a single number misleading, because 88.4 per cent on one app beside 23.2 per cent on another is two findings rather than an average. Whichever tool you buy, that discipline is what makes the percentage survive a question from a sceptical colleague.

Questions buyers ask about measuring share of voice in AI answers

Which tool gives the most accurate share of voice?

The one that asks each question most often and tells you how often it asked. On published rows that is Evertune from its Pro plan, at 100 samples a prompt.

Can I compute a share of voice without buying anything?

Yes, for a small question set. Write 20 buying questions, ask each one five times on two assistants, record which brands appear in each answer, and divide. The arithmetic is the same arithmetic; the tools buy you scale and a schedule.

What denominator should I use?

Readings is the honest default, because it is the only one of the three that you control and can repeat. Mentions and competitor sets both move when somebody else changes their marketing.

Is a market index share the same measurement?

No, and this is where Ahrefs is doing something different. Dividing against an index of 519 million prompts answers how visible your brand is in a category somebody else sampled. Dividing against your own question set answers how often your buyers hear your name. Both are legitimate and they are not interchangeable.

How many readings before the percentage is stable?

Enough that one answer changing does not move the figure a point. With 20 questions asked once, a single different answer moves your share by five points, which is why a sample count per prompt matters more than a long question list.

Two numbers, printed next to each other, settle this category

Ask any vendor on your shortlist for the readings a month and the samples a prompt, and write both on one line. Everything else in a share of voice pitch is downstream of that pair: the dashboard, the trend line, the competitor grid. Our own published counts sit beside the five above, and the sources that produced them are in the 4,947 domains that were cited and the ten everybody argues about, which is the other half of a share of voice conversation: whose pages the engines were reading when they chose somebody else.

How we know this

We read the assistants rather than guessing what they do. In our own audit of 1,260 saved answers from ChatGPT and Gemini across the United States, the United Kingdom and India, 82.0% of every citation landed on a site the business itself owned, and review directories took 0.7%. Counted as one domain per reading, 16,516 of 20,143.

That is the whole asymmetry: the pages you own are the pages an assistant quotes back to a buyer. See the dated readings and the method on our proof page.

Find out whether assistants name your business, on your own domain and your own buying questions.

By Aditya Goel, Founder, AI Knows Us. Reviewed 1 November 2026.