Share of voice in AI answers is the portion of all brand mentions or citations in AI-generated answers that belong to one brand, measured across a fixed set of prompts and compared with a defined group of competitors. If AI assistants name five brands in total when asked about a product category and one of them accounts for a quarter of those mentions, that brand has a share of voice of about 25 percent for that set of questions.
The metric comes from advertising, where share of voice meant a brand's portion of all ad spend or ad impressions in a market. In AI search it describes how much of the written answer space a brand occupies when people ask questions in its category. It is one of the main ways to express AI visibility as a comparison with competitors, where a raw mention count would be an absolute figure.
How is share of voice in AI answers calculated?
The calculation needs three inputs: a list of prompts, a list of brands to track, and the answers those prompts produce on one or more AI platforms. A prompt is the question typed into an assistant, such as "Which project management tool is best for a small agency?"
- Choose the prompts. They should match questions real customers ask, including comparison questions and questions that do not name any brand.
- Choose the competitor set. Share of voice is always relative, so the result depends on which brands are counted.
- Run each prompt on each platform and save the full answer.
- Count how many times each brand is mentioned or cited.
- Divide each brand's count by the total count for all tracked brands.
A small example: across 20 answers, brand A is mentioned 30 times, brand B 50 times and brand C 20 times. The total is 100 mentions, so A has 30 percent share of voice, B has 50 percent and C has 20 percent. Adding a fourth competitor to the set would lower every figure, even if nothing in the answers changed.
Which versions of the metric are used?
Teams count different things, and the choice changes the result. The most common versions are listed below.
| Version | What is counted | What it shows |
|---|---|---|
| Mention share | Times a brand name appears in answer text | How often the model talks about the brand |
| Answer presence share | Answers that mention the brand at least once | How many questions include the brand at all |
| Citation share | Links to the brand's own domain in the answer's sources | How often the brand's website is used as a source |
| Position-weighted share | Mentions scored by where they appear, with first place worth more | Whether the brand is recommended first or listed last |
Mention share and citation share often diverge. A brand can be named frequently because review sites and forums discuss it, while its own pages are rarely cited. The reverse also happens: a company's blog is cited as a source for a general explanation, but the answer recommends competitors. How AI search chooses which sources to cite explains why the pages a model reads and the brands it names are separate outcomes.
Some teams also tag each mention as positive, neutral or negative. A high share of voice made up of warnings about a product means something different from the same share made up of recommendations.
Why does share of voice change between runs?
An AI answer is generated fresh each time. A large language model picks its words with some randomness, so the same prompt can produce a different list of brands a minute later. The retrieval step adds more variation, because the search queries the model writes and the pages the index returns are not identical on every run.
Platforms also differ from one another. ChatGPT, Perplexity, Gemini and a Google AI Overview each run their own index, ranking and model, and some answer from training data without searching the web. A brand can hold a large share on one platform and almost none on another.
For these reasons a single run is only a sample. The number becomes meaningful when the same prompts are repeated on a schedule and the results are averaged over many answers. The article on why one AI visibility check is not enough covers this sampling problem in more detail.
How is it different from share of voice in traditional search?
In classic SEO, share of voice is usually estimated from rankings: a brand's pages hold certain positions for a set of keywords, and each position is given an expected click rate. The unit is the page, and the ranking is fairly stable from one day to the next.
In AI answers the unit is the brand name inside a paragraph. An answer either mentions the brand, cites it, or leaves it out; there is no equivalent of position ten. A brand can appear in an answer without owning any page that ranks, because the model took the name from a third-party comparison article. The opposite case exists too, where a site ranks first in the classic results and is still left out of the generated answer.
What does share of voice mean for a website?
Share of voice shows where a brand stands relative to the competitors a buyer is likely to hear about from an assistant. It does not measure visits. Clicks from AI answers are tracked separately as AI referral traffic, and the two numbers can move independently, since many people read an answer without clicking any link.
The figure is only as useful as the prompt list behind it. Prompts that name the brand inflate its share, and a competitor set that leaves out the strongest rival makes the result look better than the market. Changes in the number can only be interpreted if the prompts, platforms and competitors stay fixed over time.
When share of voice drops, the cited sources in the saved answers usually show the cause. It might be a new comparison article that omits the brand or a forum thread favoring a rival, or the brand's own pages may no longer be retrieved. The Searcherries AI visibility tracker records competitors and cited sources for each scheduled prompt run, and that is the data needed for this kind of diagnosis.
