Obility Editorial · · 4 min read
Published formulas divide by different things
Share of voice is the number most AI visibility reports lead with, and the one least often defined. The idea comes from advertising: your portion of everything said in a category. In AI answers it usually becomes your brand’s mentions divided by all brand mentions across a set of prompts. Published guides on the question do not agree on more than that. Some divide mentions by mentions. Some count the share of prompts where you appear at all. Some count citations to your domain instead of your name, and some weight a mention by where it sits in the answer.
Each of these is a reasonable measure. They are different measures, and they can move in opposite directions on the same set of answers. Before a share of voice figure goes into a board deck, someone should be able to say what was divided by what.
The competitor list sets the denominator
Take an illustrative example. You track 20 buying questions in one engine and collect one answer to each. You also track three competitors. Across the 20 answers your brand is named 6 times, and the three tracked competitors are named 12, 9 and 5 times. Two companies you do not track are named 8 and 4 times.
Divide by the tracked brands only and your share is 6 of 32, about 19 percent. Divide by every brand the answers named and it is 6 of 44, about 14 percent. Count the questions where you appear at all and it is 6 of 20, or 30 percent. The answers are identical in all three cases. Only the division changed.
The 19 percent is the figure most likely to be reported, because a tool can only count the names it was told to look for. It also hides the most useful finding in the data: two companies the answers treat as part of your category, which your team does not. A competitor list written from your sales pipeline tells you who you meet in deals. The answers tell you who buyers are shown before they reach you.
Your share can rise while nothing about you changes
Run the same 20 questions a month later. You are still named 6 times. Your largest tracked competitor drops from 12 mentions to 6, perhaps because the engine changed what it retrieves. Your tracked share goes from 19 percent to 23 percent. Nothing you did caused that, and your presence rate is still 30 percent.
The reverse happens too. A new entrant starts appearing in answers, you add it to the tracked list, and your share falls overnight without a single answer changing. A ratio moves when either side moves. Read the numerator and the denominator before you read the share, and treat any change to the competitor list as a break in the trend line rather than a point on it.
Counting rules move the number as much as the answers do
Counting every mention gives more weight to long answers. A response that lists ten vendors adds ten mentions to the denominator, while a response that names two adds two, even though the second is closer to a recommendation. Counting each brand at most once per answer avoids that, but it treats a passing mention in a list of ten the same as being the one company the answer recommends.
So a mention, a recommendation and a citation belong in separate columns. A brand can be named often and recommended rarely. A page can be cited as a source in an answer that recommends a competitor. Pooling engines causes a similar problem: a strong share in one assistant can hide a weak one in another, and a single blended figure says little about where to work.
Report the share next to the counts behind it
A share of voice figure is useful when it can be traced back to answers. Keep the competitor list fixed for the period you report, and version it when it changes. Add a bucket for every other brand the answers named, so untracked companies stay visible. Report the raw counts, the presence rate and the share together, per engine, with mentions, recommendations and citations kept apart. When two of those numbers disagree, the disagreement is usually the finding.
Obility’s Answer Engine Insights is built to be read this way. The competitor set is an input your team defines. Each observation keeps the competitors the answer recommends and the sources it cites, with the prompt wording, language, location and collection time attached. When the share moves, the team can go to the answers behind it and see whether you gained ground or a rival lost it.
Share of voice is a relative measure, so use it as one
Share of voice cannot tell you whether answers in your category are good for you. A rising share in answers that describe you as the expensive option is not progress. It is not market share either, and it does not predict pipeline on its own. Figures from two tools are only comparable if they use the same prompts, the same competitor list and the same counting rule, which is rare, so a vendor benchmark against “the industry” deserves the same question: divided by what?
Read with its counts in view, it does one job well. It shows whether the answers buyers see are moving toward you or away from you relative to the companies you compete with, and which of those companies is doing the moving.