how a score gets made.

1. 1.we take a url and pull a short excerpt of the page (roughly 400 characters) — enough to establish what the brand sells and who it’s for.
2. 2.a language model reads that excerpt and infers basic context: category, scale, likely competitors.
3. 3.from that context, we generate a set of buyer-style questions — around a dozen per brand — scoped to the brand’s category and geography. these are the kinds of questions a prospective customer might actually ask an ai assistant, not seo keywords.
4. 4.each question is sent to three engines in parallel: chatgpt, gemini, and claude. we read the raw text response — nothing is fed back in as a leading prompt.
5. 5.
   each response is scored across four dimensions:

   • —presence: is the brand mentioned at all
   • —position: how early / how prominently it appears in the answer
   • —sentiment: how favorably it’s described
   • —share of voice: how it’s mentioned relative to competitors named in the same answer
6. 6.an overall score (0–100) blends how often the brand is mentioned with the quality of those mentions. scores are floored above zero — a single ambiguous or partial mention shouldn't read identically to total invisibility, and floors also absorb model noise at the low end.

see limitations for what this methodology does not establish.