# AI share of voice: how to measure your real share of AI answers > AI share of voice shows your share of all the businesses an AI names. How to calculate it yourself, and the three mistakes that skew the result. Berger+Team · Oct 30, 2026 Topic: [Measuring visibility →](https://btlabs.dev/en/measuring-visibility) [AI share of voice](https://btlabs.dev/en/glossary/share-of-model-voice-somv) is your share of all the businesses an AI names when it answers a fixed list of customer questions. The number answers a different question from your own mention rate: not “Am I mentioned?” but “How much room do I get next to everyone else?” ## What AI share of voice means Share of voice comes from advertising planning and describes a business's share of its sector's total presence, traditionally measured in advertising spend or media coverage. Applied to AI answers, you count which businesses a model names for a question and work out what share of those mentions is yours. The same idea also goes by the name Share of Model Voice. The principle is identical: what counts is not a position in a list but the room you get in the answer the customer actually reads. ## Mention rate and share of voice are two different numbers Your mention rate tells you in how many answers you appear. Your share of voice tells you what share of all mentions you hold. The two can be far apart. Take an example: ten questions put to three models with web search give you 30 answers. You appear in 15 of them, a mention rate of 50%. But those same 30 answers name 150 businesses in total, and you only 15 times. Your share of voice is then 10%. So if you only know your own rate, you cannot tell whether you are ahead or one of many. Which metrics a small business should watch in the first place is covered in [What you measure when position one is worth nothing](https://btlabs.dev/en/posts/what-you-measure-when-position-one-is-worth-nothing). This article is about the calculation behind it. ## How to calculate your share of voice yourself You don't need a tool for this, just a fixed method and some patience. Five steps: 1. **Fix your questions.** Five to ten questions real customers ask, without your business name. Ask them in exactly the same words every time. 2. **Keep the two kinds of answers apart.** A model with web search answers differently from the same model drawing only on what it has learned. Count both separately and never add them up. In our own measurement, models with web search named us in 16 of 27 answers (run of 26 October 2026), and from model knowledge alone in 0 of 61 (run of 5 October 2026). 3. **Count every business.** For each answer, note which businesses are named, not just whether you are among them. 4. **Clean the list.** Before you calculate, three things have to come out of it; they are covered in the next section. 5. **Calculate per model.** Your mentions divided by all business mentions, times 100\. And across several runs, not from a single one: answers change from week to week, as [GEO by model: why every AI search answers differently](https://btlabs.dev/en/posts/geo-by-model) shows. ## Three mistakes that skew your share The formula is simple. The hard part is deciding what goes into the numerator and what into the denominator. ### Your own brand ends up on the wrong side In the list of domains our citation monitor counts in the answers, berger.team sits at the top with 120 mentions (as of 30 October 2026). That is not a competitor but the website of Berger+Team; btlabs.dev is the website of the product. If you don't merge company name, product name and all your own domains into a single sender, you make yourself look smaller and everyone else bigger. ### Sources are not competitors Among the most frequently named domains in our data are also an OpenAI help page and a chamber of commerce. They are cited as sources but recommend no business. Only businesses that offer what you offer belong in the denominator. ### Mentioned is not recommended An answer can mention you in passing, recommend you or name you as the first choice. If you count all of them the same, you overestimate your share. Count the three levels separately or weight them, for example one point for a mention, two for a recommendation, three for first choice. Which weighting you choose matters less than keeping it the same across all runs. ## What the number cannot tell you A share of voice from ten questions is a sample, not market research. It shows that something is moving, not why an AI names someone. Then there is the way it is measured: our monitor queries the models through an intermediary service, not through the apps your customers use. What they see there can differ. And the number stays honest even when it is uncomfortable. Asked “How can a business measure whether it is visible in AI answers?”, none of the three models with web search named us in the run of 26 October 2026. ## How often to measure With web search, one run a week is enough, because these answers react to new content within days. What a model has learned changes over months, so one run a month is sufficient there. You only see a direction across several runs, not between two weeks. How your share in AI answers fits together with visitors, AI crawls and search data is shown in [Measuring visibility](https://btlabs.dev/en/measuring-visibility). ## Frequently asked questions. **How do I measure whether my business appears in AI answers?** The most direct way: test it yourself. Ask the popular AI assistants the questions your customers would ask — for example, “which business near me does…?” — and check whether your business is mentioned and whether the details are correct. Repeat this regularly and in several languages, because answers change. In addition, website statistics show whether AI systems access your content: those visits leave traces. So that it does not stop at spot checks, btlabs Core has a visibility cockpit: it brings search data and AI systems' visits to your website together in one place, presented understandably. You see trends instead of snapshots — and notice early when details about your business need correcting. **Do I need a tool to measure my share of AI answers?** Not to start with: a spreadsheet will do. Ask five to ten customer questions without your business name, always in the same words, note every business named in each answer and work out your share per model. A tool mainly saves time and keeps your questions identical over months. If you use one, check that it counts answers with web search and answers from model knowledge separately, and that it merges your own brands and domains into a single sender. **How often should I measure my mentions in AI answers?** With web search once a week, for model knowledge once a month. Answers with web search react to new content within days and vary from week to week; what a model has stored only changes over months. That is why you only see a direction across several runs, not between two weeks. Next step ## Want to look at it together? No pitch, no standard package — an honest take on your project. A reply, usually within 24h. [Start a conversation →](https://btlabs.dev/en/contact#cta) ## You might also like - [· What you measure when position one is worth nothing · Average position is losing its meaning, because an answer nobody clicks has moved in between the ranking and the customer. What replaces it, how to measure that — and the trap where more impressions at unchanged clicks look like progress.](https://btlabs.dev/en/posts/what-you-measure-when-position-one-is-worth-nothing) - [· How do you check if an AI is actually citing you? · A simple self-check shows whether ChatGPT and Perplexity recommend your business correctly — and why checking once is not enough.](https://btlabs.dev/en/posts/does-ai-really-cite-me-self-check) - [· From Inquiry to Customer: How an AI Visitor Really Converts · The AI recommendation is only half the job. What happens between "the AI names your business" and "the inquiry actually arrives" — with a real South Tyrol angle and three languages.](https://btlabs.dev/en/posts/from-ai-visitor-to-customer-south-tyrol) --- Source: https://btlabs.dev/en/posts/ai-share-of-voice Languages: [de](https://btlabs.dev/llms/de/posts/share-of-voice-ki-antworten-messen) · [it](https://btlabs.dev/llms/it/posts/share-of-voice-risposte-ia) See also: [llms.txt](https://btlabs.dev/llms.txt) · [ai.txt (Policy)](https://btlabs.dev/ai.txt) · [identity.json](https://btlabs.dev/identity.json)