A percentage is useful only when backed by consistent prompts, complete responses and transparent counting rules.
To assess GEO, I preserve the exact questions, AI interfaces, dates and original responses. Without these, a percentage can be misleading.
BMR is the share of received AI answers that mention the target brand. Share of Voice is the brand's share of mentions among a specified set of competitors. These are different metrics with different denominators.
A brand mention is not necessarily a positive recommendation. I also review whether AI proposes the company for the actual buying need, its position in the answer and the cited sources.
BMR = responses mentioning the brand / all received responses × 100%. Multiple appearances in one answer count once. Prompts already containing the brand should be separated from non-branded questions.
Hypothetical example: 25 responses mentioning the brand out of 100 give BMR 25%. A comparable follow-up retains the same questions and AI interfaces.
Expanding from 40 questions to 52 changes the sample. I preserve the original tracking series and measure new topics separately rather than comparing 2 different totals as one trend.
Ten mentions in 20 answers give 50%. Ten mentions in only 18 returned answers give 55.6%, even though the number of positive answers did not increase. Both expected and received totals should be reported.
A domain in an AI source list does not necessarily mean the company was recommended. Automated brand matching needs checking, particularly for ambiguous names.
Across the same 6 prompts and 7 interfaces, MethodLab BMR increased from 19.0% to 52.4%. Three platforms reached 6 of 6 mentions while Yandex Search with Alice remained at zero. Platform-level figures explain the aggregate.
Read the industry GEO guides for business-specific examples.
I will review commercial prompts, competitors and the sources used by ChatGPT, Alice, Google and other AI services.
Get a mini-audit