What we measure. What we don’t.

Your sample

You confirm competitors and buyer prompts. Each run records engines, provider/model, time, and US English locale. API outputs differ from consumer ChatGPT or Gemini experiences and do not reproduce individual personalization. Models and web sources change; repeated runs may differ.

Salience and sources

Not mentioned: no matched brand alias. Mentioned: a matched neutral appearance. Cited: a source belongs to the brand. Recommended: a matched appearance classified as a positive choice. Unknown: unavailable, failed, or ambiguous classification. An unavailable citation list is different from zero citations.

Deterministic matching prevents the classifier from inventing a brand mention. Aliases and classifiers can still make mistakes; inspect the answer behind the result.

Metric definitions

Tracked visibility is the share of successful answers where the brand is mentioned, cited or recommended. Recommendation share counts confirmed recommended answers among successful answers. Owned citation share divides owned citations by all observed citations. Successful answers with unknown classification remain in the denominator but never count as confirmed positive results. Their count is shown explicitly; a zero confirmed share does not establish absence when answers remain unknown. Failed calls are excluded from the denominator. Comparisons use the same controlled sample and confirmed competitor version.

Diagnoses are hypotheses

Opportunities connect observed answers with content and evidence gaps. Confidence reflects supporting evidence, not a guarantee that a change will influence an AI model. Priority combines commercial intent, recurrence, competitor advantage, evidence strength, confidence and effort.

Supported WordPress content patches require an exact-version approval, a verified conditional connector, current permissions and live verification. Exported drafts and customer-reported implementations remain separate from Seevaro-verified changes. Retests may be inconclusive and do not prove causality.

Engine coverage

Only configured supported APIs are sampled; a run may use one available surface. OpenAI Search API and Gemini Grounded API adapters exist. Perplexity, Claude (Anthropic web search) and Grok (xAI web search) are added only after verification, in that order, up to your plan's engine limit. Coverage is enabled only after configuration and live acceptance. Official APIs are used; consumer interfaces are not scraped. Google grounding may return redirect URLs, whose final ownership cannot be inferred safely without resolution.