How we compute AI-search visibility

Each tracked prompt is run against a battery of LLMs — Claude / GPT / Gemini via OpenRouter — on the cadence you set (daily / weekly / monthly). Every (prompt × model × run) is one observation.

Brand cite-rate = observations where your normalized_host appeared (case-insensitive) in the response, divided by total observations. Target-URL cite-rate is the same but requires a URL match, not just a brand name.

Competitor + sentiment

With `aeo_competitor_layer_enabled=true`, each response gets a second pass: competitor names/domains from `project.competitors` are matched lexically, and the brand mention's sentiment is classified with a one-shot LLM call. Adds ~$0.01 per prompt-model run.