Mention rate
The fraction of responses that mention the brand. If 200 prompts produce 600 responses (across engines and repeats), and 60 of those responses mention the brand, mention rate is 10%. Mention rate is the rawest visibility signal. A low mention rate does not always mean poor visibility — it may mean the prompts do not naturally surface the brand’s category. The platform tracks this through thevisibility metric, weighted by engine usage proxy in the score.
Share of voice
The brand’s mentions divided by the sum of all tracked brand and competitor mentions. If the brand is mentioned 60 times and competitors are mentioned 140 times, the brand’s share of voice is 30%. Share of voice is competitive. It answers: of all the answer-engine conversation about this space, how much belongs to the tracked brand? This metric has the highest weight (0.35) in the GEO Score because it directly measures competitive positioning. The intuition: a brand that improves from 10% to 20% share of voice has doubled its relative presence, even if absolute mention rates are low. A brand with high mention rate but low share of voice is frequently mentioned alongside dominant competitors. Share of voice is calculated per-engine and combined through weighted mean using the engine weights fromengines.yaml.
Position
Where in the answer the brand first appears. First position means the brand is the first named option or entity in the response. Position is inverted for scoring — earlier appearance gives a higher score. The platform records average position per engine. Position contributes 0.15 weight to the GEO score. An improvement from position four to position two is meaningful even if mention rate stays flat.Sentiment
The measured tone around brand mentions, normalized to a scale from negative to positive. Raw sentiment values from the extractor are normalized into the 0–100 range for scoring. Sentiment has the lowest weight (0.10) in the GEO score. This is intentional. Sentiment signals from LLM responses are noisy — a neutral factual description (“X was founded in 2020”) scores differently from an evaluative one (“X is the market leader”). The weight reflects the lower reliability of automated sentiment classification for LLM text. The platform tracks sentiment as a directional indicator. A negative shift warrants investigation. A single negative mention among many neutral ones is not a pattern.Citations
Whether the engine attributed a brand fact to a source. A citation is recorded per-engine when the response includes a reference (inline number, URL, footnote) that points to brand-controlled content. Citation rate is not a direct score component but feeds the evidence behind recommendations. A brand with high mention rate but low citation rate appears in answers without attributable sources — the engine is saying things about the brand without linking to brand content.Confidence intervals and sample size
Every metric has ann value — the number of observations that produced it. A visibility score of 40% with n=500 is more reliable than 40% with n=5. The platform reports n alongside every component score.
The GEO score trend includes n per data point. When reading a score change:
- Low score, high
n: The engines consistently do not mention the brand. The result is reliable. Focus on retrieval and content gaps. - High score, low
n: The brand appeared in a small sample. The result may not generalise. Run more prompts or increase the preset size. - Low score, low
n: Not enough evidence to conclude anything. Run a larger measurement.
How metrics become the score
The five score components and their weights transform raw metrics into the 0–100 GEO score:
Each component is normalized to 0–100 before weighting. Missing components drop their weight and the remaining weights renormalize. See GEO Score for the exact formula.
See also: Reading your results for practical interpretation.