How we calculate the numbers
We measure what AI models answer to your customers’ real questions. Here’s exactly how — and what we don’t do.
Last updated: November 11, 2026
What we measure in AI answers
| Metric | How it’s calculated | Keep in mind |
|---|---|---|
| Answers with your brand | Answers where the brand appears ÷ successful answers | The brand is named in the answer text: the name, the domain or a variant of the name. Failed or not yet analysed answers are not counted. |
| Answers with a recommendation | Answers where AI recommends the brand ÷ successful answers | A mention is not necessarily a recommendation. |
| Answers linking to your site | Answers with at least one link to your domain ÷ successful answers | Source links count here, not as mentions. Subdomains of your site count. |
| Share among the tracked brands | Brand appearances ÷ appearances of all tracked brands | One appearance per answer. It is not a market share. |
| Position | Average place in the explicit list of recommendations in the answer | No list — “undetermined”. Brand absent — “not present”, not 0. |
Under 10 answers we show “small sample” and draw no conclusions. A mention in the text and a source link are different things: we count them separately.
How we run
- We ask the models through their official APIs, with web search on where the model offers it, from the city or country of the business.
- With a subscription, every active question is measured weekly.
- In Starter, the site audit is redone every 30 days.
- In Business and Agency, the site audit is redone every 7 days.
- The free check is a single measurement: 3 questions in ChatGPT and Gemini, with no history.
- We show a change as a change only when it exceeds the usual variation between runs.
The GEO Readiness score
Methodology version: geo-1.5
Shows how ready the checked pages are for AI to find and cite information about the business. It is not the probability of appearing in answers.
| Content accessibility | 20% |
|---|---|
| Structure and navigation | 15% |
| Coverage of customer questions | 25% |
| Clarity of business facts | 15% |
| Credibility and freshness | 15% |
| Language and regional consistency | 10% |
- Checks that do not apply are excluded from the score; they mean neither a pass nor zero.
- Checks not done lower the coverage but do not count as zero.
- A category counts in the score only with a coverage of at least 50%.
- Coverage of at least 70% — final score; 40–70% — preliminary score; under 40% — the score is not shown.
- The free check reads up to 3 pages, which is why it says “checked on 3 pages”, not “site audit”.
The general site audit
Methodology version: site-1.5
A separate methodology with its own score: it is not mixed with GEO Readiness or with the presence in AI.
| Technical accessibility | 20% |
|---|---|
| Business clarity | 15% |
| Service pages | 20% |
| Content quality | 15% |
| Trust | 10% |
| Contact | 10% |
| Mobile & performance | 10% |
Geographic data sources
We identify cities using GeoNames data, normalized to search for alternative names. GeoNames · CC BY 4.0.
We measure through the API, not the ChatGPT interface
- Apps use your history, account and location; the API has no access to them.
- Models update often, and the same prompt can get different answers from run to run.
- That’s why we track the weekly trend, not a single answer.
What we don’t do
- We don’t invent search volumes or prompt frequencies.
- We don’t guarantee appearing in AI answers.
- We don’t change your site — you decide what’s published.
- We don’t use your project data to train models.
Your data
- AI answers are kept 12 months; account data until the account is deleted.
- You can export data anytime from Settings.
The score doesn’t depend on BRTeam’s services: the rules are the same for everyone. About us →