Prompt-level Visibility

AI assistants build each answer from scratch, so there is no fixed “page one” to rank on. Ask the same question two slightly different ways and you can get two different answers, with a different set of brands named each time. That is why measurement moves down to the single prompt: you run a defined prompt, check whether your brand appears, and note the role it plays (named leader, listed alternative, or buried footnote).

What a prompt tracker records

Most programs group prompts into three buckets:

Branded:     "What does Acme do?"
Category:    "best warehouse robotics vendors"
Competitor:  "Acme vs Rival for mid-market"

Category prompts carry the most weight, since that is where discovery-stage buyers form a shortlist before they know your name. Because output varies run to run, you sample each prompt several times per engine and track presence as a rate, not a yes/no. Individual prompts then roll up into share of voice (how often you are named versus rivals) and prompt penetration (how many of your tracked prompts mention you at all).

Presence is engine-specific. You can dominate ChatGPT and be absent from Perplexity or Gemini, because each pulls from different sources, so a real program tracks every engine your buyers use. The academic groundwork here is the GEO paper, which framed content “visibility” inside generated answers as a metric you can measure and move.