GEO probe methodology
Page scores answer whether models can read you. Probes answer whether models will say you.
Page score vs probe score
Page GEO (Princeton KDD 2024 heuristics, Schema, llms.txt) measures input citability. D/K probes measure output: named recognition vs unprompted mention. Do not weight them into one score.
Mention vs recommendation
Mention: the brand or official host appears. Recommendation: structured relation=positive, i.e. an explicit preferred option. Being named without a recommendation must keep recommendation at 0.
Recognition vs mention
Recognition comes from protocol D (domain-only prompt). Mention comes from protocol K (keyword-only prompt). Being able to describe a named domain is not the same as bringing it up unprompted.
Citations and content writeback
Citation rate: whether URLs in the answer match the official host. Content writeback: whether those URLs equal a published_url you shipped. Only the latter shows that a specific piece of shipped content was used.
Same-condition retest
Freeze aliases, keywords, and competitors. After page fixes or publishing, run the same watchset again. Compare in percentage points and keep the previous original answers.
No single AI-visibility score
Recognition, mention, recommendation, and citation have different denominators and prompts. A blended score hides 'recognized but not recommended' or 'mentioned but your article was not cited'.