AI visibility numbers are synthetic and non-deterministic — the same prompt can return a different answer twice. Most teams sample with a prompt list built in an afternoon, then hand the result to an executive with no reason to believe it. The fix starts in the unglamorous part of the stack: a prompt index built on real topics, personas, intents, and branded ratio, extended with MCP and an LLM into analysis that used to need a full data team, distilled into a short list of barometers leadership can actually plan against: share of voice, recommendation quality, narrative risk, and the sources driving them.

Spencer will walk through the whole stack, from building an index that holds up to mapping the methodology back to business goals. Tool-agnostic; bring your current prompt set.