Knowledge base

GEO monitoring metrics

Track mentions, citations, position, sentiment and competitor co-occurrence with fixed prompts and a fixed model set to build a comparable series.

GEO monitoring does not chase rankings. It tracks five things in model answers: whether the brand is mentioned, whether a page is cited, where it appears, how it is described, and which competitors appear beside it. A ranking can rise while citations stay flat, and the reverse also happens. Fix those five and results from different dates can be compared.

Keep the prompt set and the model set fixed. Same questions, same models, same recorded fields; only the date changes. Prompt edits, model version upgrades and sampling changes must be flagged separately, or the series stops being comparable. If a provider retires a model, record the replacement and start a new series instead of splicing the two together.

The five weights are fixed: AI visibility 30, AI citability 25, search visibility 20, technical foundation 15, brand trust 10. When no model test has run, AI visibility is marked not computed and removed from the total instead of scoring zero, so missing data never punishes the brand. The same rule applies to any dimension whose data source is unavailable.

Coverage has three bands. At 0.80 or above the total is published; from 0.60 to 0.79 it is published with a note that some dimensions lack data; below 0.60 only dimension scores and the gap list appear. Missing values are marked unavailable, never treated as a pass. A pass requires evidence; an empty field never counts in your favour.

Review in order: access, answer-first structure, entity evidence. Record mentions and citations once a month and keep the evidence table from that run, because a score without its evidence cannot be checked later. Metrics describe what changed; they do not promise that any edit will be recommended, or that a competitor will lose ground.