Methodology · Visibility

How the GEO share of voice (AI answers) is calculated

Method version geo-v1 · Last updated 2026-10-13

In short

GEO (generative engine optimisation) share of voice is a brand's share of the mentions that tracked brands receive in AI assistant answers to a fixed set of buyer questions. Every month each question is asked twice to Claude with live web search; a mention counts 1 point and a citation of the brand's own website counts 0.5 extra. Results are also shown per topic and per question.

Formula

Points of brand b = Σ mentions of b + 0.5 × Σ citations of b's own site
GEO share of b = points of b ÷ Σ points of all tracked brands × 100
Per question: a brand counts as named when any of the two answers names it, ordered by its average position

How it is calculated, step by step

Take the questions from the market profile: per topic and language, questions a buyer would ask, without brand names.
Ask each question twice to Claude with web search, localised to the market.
Read out every brand named in each answer, in order of appearance, and every website cited as a source.
Match the names to the tracked brands (the rest is the rest of the market) and the cited URLs to the brands' domains.
Score points per brand and divide by the total; repeat per topic for 'Where each brand wins'.

What goes into it

Questions: Non-branded buyer questions per topic and language, editable in the market profile.
Answers: Two sampled answers per question, because AI answers vary.
Mentions and citations: Brands named, their order, and sources cited.

Data sources

Claude Sonnet with web search (answers)
Claude Haiku (reading the brands out of each answer)

How often it is updated

Automatically on the 1st of every month, and on demand (at most once a week).

How to read it

Topics you lead show where assistants already associate you with what buyers want; topics where competitors lead are your GEO priorities.
Named first matters: users often act on the first recommendation.

Example

Named in 6 answers and cited twice gives 6 + 0.5 × 2 = 7 points; with 50 points for all brands together that is a 14% share.

Limitations

AI answers vary between runs; two samples per question reduce but do not remove this.
One assistant (Claude) is measured today; others may answer differently.
Results depend on the question set: changing the questions starts a new version.

Frequently asked questions

Why don't the questions mention brands?

A question that names a brand gets that brand mentioned by default. Open questions measure whom the assistant recommends unprompted.

Why ask each question twice?

Language models give different answers to the same question; two samples make the result more stable.

AI readiness score — How well AI assistants can reach and use a site: 60% bot access, 30% content readiness (assessed by Claude), 10% llms.txt.
Rest of the market — The part of the clicks or AI mentions that goes to sites and brands you don't track: comparison sites, media, forums and other providers.
Share of voice (Visibility) — Your share of the visibility of all tracked brands together, per channel (SEO, GEO, backlinks, socials); the Visibility score is the average of the four.