A branded prompt contains your business name or a tracked competitor’s name (e.g. “Is Stripe better than Adyen?”). LLMs trivially cite a brand when the brand is in the question — those numbers flatter your visibility score.
The unbranded number is the real signal
The Visibility tab isolates the unbranded number — buyer-discovery questions that don’t mention you or your competitors. That’s the score that actually reflects whether you’re winning at AI answer-engine optimisation. If your unbranded number is 0% but branded is 90%, the engines know who you are once asked but don’t recommend you to buyers searching for the problem you solve.
Where the classification happens
Step B in our pipeline classifies each generated prompt as BRANDED, UNBRANDED, or UNKNOWN before any probe runs. Legacy prompts predating the classifier are tagged UNKNOWN; new prompts always get one of the first two.
Mixing the two
The Visibility tab toggles between unbranded-only, branded-only, and all-prompts views. The Run Detail page shows both breakdowns side-by-side. Most workspaces watch unbranded movement weekly and treat branded as a sanity check (if branded starts dropping, something’s wrong with your homepage schema or your name is being confused).