How to read share of voice
Dylan Ander
Share of voice, in AI search, is a fraction. Count the answers in a fixed prompt set where your brand is named. Divide by the number of answers. That is the number. Everything else people do with it is interpretation, and most of the interpretation is sloppy.
I like the fraction when the denominator stays honest. I dislike what happens when the denominator is hidden and the percentage gets a logo.
Named is not recommended
A mention means the model said your name. A citation means it pointed at a URL, yours or someone else's, as a source. Rank is where you sat when the answer was a list. Sentiment is the tone wrapped around your name. Share of voice is how often you appeared at all.
These are different facts. I have watched a brand get mentioned in almost every answer and recommended in almost none. The model knew the name and would not trust it with the sale. If you only chart share of voice, you will celebrate being famous for the wrong reason.
When a report is doing its job, the columns sit next to each other.
| Column | What it is actually telling you |
|---|---|
| Mention rate | The model said your name. |
| Citation rate | The model pointed at a source. |
| Rank | Where you sat when it made a list. |
| Sentiment | The tone wrapped around your name. |
| Share of voice | How often you appeared at all, across the fixed prompt set. |
If a report gives you one blended score, ask them to unblend it. If they cannot, the score is a costume.
The denominator is the whole game
Share of voice moves when you change the prompts, even if every model kept the same opinion of you. Add ten prompts you already win and the line goes up. Drop the prompts a competitor owns and the line goes up. Neither event means you gained visibility. It means you edited the test.
So the first question I ask is: show me the prompts. The second is whether that list stayed fixed long enough to compare weeks. A single afternoon is an anecdote. Direction across the same prompts is closer to a measurement. If you have not written the list yet, start with how prompts work.
Engines do not vote as a bloc. ChatGPT can name you while AI Overviews cite a magazine that never mentions you. Averaging them into one pretty number hides the only useful part, which is the split.
How I refuse to read it
I do not treat a dashboard percentage as the market. I treat it as a claim about a list, a day, and a set of engines. If any of those three is missing, I ignore the claim.
This site is not that dashboard. The sign-in on the tracker page is a placeholder, on purpose, until an account exists. I would rather show you how to read the number than hand you a figure and ask you to trust it.
One habit is enough. Never accept share of voice unless the prompt list is in the same breath. After that, look at where the answers are actually coming from. A lot of the time, the source is a thread. That is the subject of how Reddit moves AI rankings.