AI share of voice: three numbers people mix up
Two teams can measure the same brand in the same week and report different share of voice figures. Usually both are right, because they divide by different things. Here is what each number counts, and how to report one that somebody else can repeat.
Figures on this page come from one dated measurement: 12 buyer questions about SEO agencies in Poland, six models through their APIs, 15 repeats each, 1,080 answers collected on 7 September 2026.
The same brand, three honest numbers
The measurement held 1,080 answers, 6,245 company names and 89 different companies. An answer usually names several companies, so the totals differ from the number of answers.
| Company | Share of answers | Share of all names | Named first |
|---|---|---|---|
| Delante | 64.2% | 11.1% | 38.1% |
| Widoczni | 54.5% | 9.4% | 6.6% |
| Whites | 24.5% | 4.2% | 3.0% |
Share of answers divides by 1,080 answers. Share of all names divides by 6,245 mentions. Named first divides by 1,080 answers and counts only the company that opens the answer.
Widoczni is named in more than half of all answers and opens the answer in one case in fifteen. A report that carries only the first column and a report that carries only the third column describe the same brand very differently.
How each number is counted
- Share of answers. Answers that contain your name, divided by all answers collected. A company counts once per answer, however many times it is written there.
- Share of all names. Your mentions divided by every company mention in the set. This one falls as soon as assistants name more companies per answer, even when your own presence has not changed.
- Named first. Answers where your company is the first one written, divided by all answers.
Citation share is a fourth number and a different thing: it counts how often your website is cited as a source. An assistant can name a company without citing it. Definitions: citation share, answer share, AI visibility.
The same split across eight markets
Each market was measured on its own questions in its own language, with 1,080 answers. The leader's share of answers and its share of all company names move apart as soon as assistants name more companies per answer.
| Market | Most-named company | Share of answers | Share of all names | Companies named |
|---|---|---|---|---|
| Poland | Delante | 64.2% | 11.3% | 88 |
| Estonia | Convertal | 35.3% | 11.6% | 141 |
| Finland | Suomen Digimarkkinointi | 39.2% | 8.6% | 178 |
| Sweden | Pineberry | 62.4% | 14.0% | 172 |
| Norway | Synlighet | 52.7% | 10.6% | 167 |
| Denmark | Bonzer | 42.1% | 9.2% | 150 |
| Latvia | SEO.lv | 32.9% | 11.4% | 148 |
| Lithuania | Market Rats | 26.1% | 7.1% | 154 |
Sweden shows the widest gap in the set: the most-named company is in 62.4% of answers and holds 14.0% of all company names, because Swedish answers name many companies each. Local-language questions, six models, 15 repeats each, 1,080 answers per market, collected 7 and 8 September 2026. These are counts of names in AI answers. They do not measure quality of work or market share. The full ten per market are on the answer engine optimization page.
How precise is the number
Less precise than one decimal place suggests. On a single question with 90 answers, one company was named in 28.9% of them. The range around that result runs from about 20.5% to 39.0%. Two reports inside that range describe the same situation.
So put the number of answers next to every percentage. A figure from 90 answers and a figure from 1,080 answers carry different weight, and a reader cannot see that from the percentage alone.
What moves the number
- Which assistants you ask. One model in this measurement named the leading company in all 180 of its answers. Another named it in four.
- Web search. The assistant that searched the web produced a different list from the five that answered from what they already hold.
- How many times you ask. New company names were still appearing at repeat 13, 14 and 15 in 42 of 72 question and model pairs.
- The language of the question. The same question in two languages can return different companies.
The method side is on the companion page: answer engine optimization and how to measure it.
How to report it
One line under every figure, with these six facts: the question, the language, the models, whether web search was on, the number of answers, and the date. With that line a colleague can repeat the measurement and compare it. Without it, the next report measures something else.
Questions people ask
What is AI share of voice?
AI share of voice is your part of the attention inside AI answers. It is usually one of three numbers: the share of answers that name you, your share of all company names mentioned, or the share of answers where you are named first.
Why do two reports show different share of voice numbers?
Because they divide by different things. One brand in our September 2026 measurement was named in 24.5% of answers and held 4.2% of all company names mentioned. Both numbers are correct and they describe different questions, so a report has to say which one it used.
What is the difference between share of voice and citation share?
Share of voice counts how often your brand name appears in answers. Citation share counts how often your website is cited as a source of an answer. An assistant can name a company without citing it, and cite a page without naming the company.
What is a good AI share of voice?
There is no single good number, because it depends on the category, the questions and the assistants you measure. The leader in our measurement of one category was named in 64.2% of 1,080 answers, while the tenth company was named in 14.5%. Compare a brand against its own earlier measurement on the same protocol.
How precise is a share of voice number?
Less precise than it looks. With 90 answers, a result of 28.9% sits in a range of roughly 20.5% to 39.0%. Report the number of answers next to the percentage, so a reader can see how much movement is normal.
How often should AI share of voice be measured?
On a fixed protocol that you repeat. Model versions change, and web search brings in pages that were published yesterday, so a figure without a date, a model and a language cannot be compared with the next one.
Measure your own share
Ask one buyer question across several models and count the answers that name you. The Dice Roller is free and needs no account.
Open the Dice Roller Browse the glossary