Common questions about measuring how AI models describe sports teams, athletes, leagues and brands — followed by definitions of the terms Bowstone uses. For the full scoring detail behind these answers, see the methodology.
Jump to questions or the glossary.
Bowstone sends a fixed battery of sports questions to four AI models every week and records which tracked entities each answer names. Because the same questions run on the same schedule, the counts are comparable week to week rather than being a one-off snapshot. No prompt ever names a brand — the point is to observe which entities the models surface on their own. Coverage currently spans 351 teams, players, leagues and brands.
Bowstone does, across Claude, GPT-4o, Gemini and Perplexity. Each model receives the same prompts, and every answer is parsed for which tracked teams, players, leagues and brands it mentioned, where in the list they appeared, and how strongly the answer recommended them. Results are stored per model rather than blended, so any figure on the site can be broken down by platform.
AI Visibility is an entity's share of all tracked mentions in a week, plus a bonus for appearing across several models rather than being concentrated on one. It is normalized to a 0–100 scale against a comparison pool, so the number expresses a position relative to peers rather than a count of mentions. AI Visibility is the largest component of the Bowstone Index, carrying 65% of the weight.
Ask the same questions repeatedly and record what comes back, because a single answer from a single model is not evidence of anything. Bowstone runs five categories of prompt — top teams, top players, brand, value and news — every week across four models, so an entity's presence is measured across question types as well as platforms. Each entity also carries a confidence indicator derived from how repeatable the prompts behind its score are, which tells you whether a week's movement is meaningful or just model variation.
Yes. Every mention is stored against the model that produced it, so each entity page shows a per-model breakdown rather than one blended figure. Bowstone also treats disagreement between models as a signal in its own right: when one model mentions an entity more than three times as often as the average of the others, that entity is flagged for model divergence.
The Bowstone dashboard ranks tracked entities by their index score, and each entity page shows its own mention counts by model and by prompt. Rankings are always drawn within a comparison pool — players against players, teams against teams — because raw counts across types mostly reflect which categories models get asked about, not which brands are strongest.
AI Mention Share is that measure: an entity's mentions expressed as a proportion of all tracked mentions in the same weekly run. Because it is a share rather than a count, an entity's number can move when its rivals move even if its own coverage is unchanged. Shares are only comparable across weeks when the prompt battery and the tracked roster are stable, so Bowstone records methodology changes as explicit boundaries in the data.
Each entity page includes Visibility Drivers, which breaks the week's movement into which prompt categories gained or lost mentions, which model moved most, and which cited sources appeared this week that did not appear last week. These are ranked by observed change, and Bowstone states plainly that this is association rather than cause — the panel shows what moved alongside a score, not what caused it.
Yes, and keeping them separable is deliberate. The Bowstone Index combines AI Visibility at 65% with Social Presence at 35%, where social is measured as week-over-week follower growth rather than audience size. Franchise valuation is tracked as Market Signal and is deliberately excluded from the index, so a gap between what a franchise is worth and how visible it is in AI answers stays visible instead of being averaged away.
By comparing movement against a baseline fixed before the change, which is why Bowstone stores an entity's index score at the moment a related article is published rather than recomputing it later. Dated events are recorded alongside the score history so a before-and-after can be read directly. One caveat matters: sponsorship activity is not itself an index signal, so a deal registers only if it changes what AI models actually say — and a major announcement producing no visibility change is itself a finding.
The vocabulary used across the dashboard, entity pages and the weekly newsletter.