Established AI Visibility
Weekly Core measures durable recognition using a frozen ranked-question battery and a published observation record.
METHODOLOGY / PUBLISHED CORE v2
Measure the answer.
Preserve what it means.
Bowstone uses separate instruments to study durable AI recognition, recent prominence, open-ended discovery and future regional variation. Each has its own population, protocol and limits.
Current published Core observation · October 31, 2026Versioned sports AI intelligence
WHAT BOWSTONE MEASURES
AI answers create a representation worth measuring.
AI systems name, rank and describe sports identities in direct answers. Bowstone observes which entities appear, where they appear and how consistently they appear across a fixed set of questions and providers.
This evidence complements social, media, performance and valuation data. It does not replace those measures or turn model output into objective truth.
01
Four pillars. Four different jobs.
Maturity labels describe what is published now, what remains private and what is still planned.
Recent AI Visibility
A separate NFL pilot measures prominence within a fixed rolling 28-day window among entities already eligible in Core.
Discovery
Current Relevance v1 uses open-ended NFL questions to surface teams and athletes, including verified athletes outside Core. No official public ranking has been released.
Regional Intelligence
A planned instrument will study differences between national recognition and recognition in specific markets. Pittsburgh and the Steelers are an initial research example, not a published result.
02
Established AI Visibility is a frozen weekly benchmark.
Core repeats one versioned instrument so compatible observations can be compared over time.
87Ranked prompts
4Equally weighted AI systems
5Team and player concepts
WeeklyObservation cycle
Recognition
How readily an entity appears among leading teams or athletes.
Brand strength
How strongly models place its overall brand or personal brand.
Commercial attractiveness
How models rank sponsorship or commercial-brand appeal.
Cultural relevance
How prominently the entity appears beyond competition alone.
Global recognition
How models rank recognition among audiences outside North America.
DIRECT RANKED ANSWERS
Position determines credit.
For each top-five question, first through fifth receive 5, 4, 3, 2 and 1 credits. A model contributes one quarter of each concept score. Incidental mentions in explanations do not earn credit.
- Rank 15credits
- Rank 24credits
- Rank 33credits
- Rank 42credits
- Rank 51credits
ONE CONCEPT
100 × earned rank credit ÷ maximum rank credit
That percentage is averaged equally across the four accepted provider answers for the concept.
TEAMS AND PLAYERS / AI VISIBILITY
70% own-cohort mean + 30% cross-sport mean
Each mean contains the same five equally weighted concepts. Own-cohort and cross-sport questions are kept distinct before the final blend.
Complete zero versus incomplete
An eligible entity absent from an accepted answer receives zero credit for that answer. A fully accepted battery can therefore produce a real observed zero.
A missing provider response makes the observation incomplete. It is not converted to zero, and the incomplete observation cannot be published as complete.
The comparison population matters
Each observation freezes a response-valid universe and a scoring roster. Team and player rankings are calculated within their compatible type and cohort.
Valid outside-roster answers are preserved as evidence without silently adding a scored entity. Public player leaderboards require at least ten score-eligible players in the peer group.
How leagues and conferences differ
Professional leagues use five equally weighted concepts and top-three rank credit of 3, 2 and 1. College conferences use recognition and brand strength, weighted equally. Current Core does not calculate Social Presence or BX for leagues and conferences.
04
Discovery asks who is surfacing now.
Current Relevance v1 is a separate, open-ended NFL instrument.
CURRENT RELEVANCE v1
Open-ended discovery across a rolling 28-day window.
- 12
- scored prompts
- 4
- providers
- 48
- required scored responses
- 4
- non-scoring shadow responses
What it records
Six team and six player prompts ask about current importance, increased relevance, reputation divergence, changed perception, material events and new storylines during exact inclusive dates.
Top-five selections receive 5 to 1 points. Rankings also report selection count and provider breadth. Shadow responses are stored but never enter scores, denominators or completion.
Discovery can extend beyond Core
Original labels, prompts, ranks, explanations and provider evidence are preserved. Verified athletes outside Core may be ranked in Discovery without being added to the frozen Core registry.
Aliases are recorded explicitly. Ambiguous, out-of-scope and unverified identities remain separate. Unresolved scored labels block formal completion rather than being silently merged.
A zero-selection result means no selection in the completed Discovery battery. It does not mean zero real-world relevance. Model descriptions of events remain model-produced evidence until independently reported and verified.
05
Recent AI Visibility measures short-window prominence.
The private v0.1 NFL pilot adapts Core's five concepts to a fixed rolling 28-day window.
RECENT AI VISIBILITY v0.1
A frozen Core population, measured through a recent-time lens.
- 20
- ranked prompts
- 4
- providers
- 80
- required accepted responses
- 28
- inclusive calendar days
Same dimensions, distinct construct
Five NFL cohort and five cross-sport questions for teams are paired with the same structure for players. Recognition, brand prominence, commercial discussion, cultural prominence and global recognition are framed around exact dates.
The pilot preserves equal provider and concept weights, 5-to-1 rank credit, and the 70% NFL cohort / 30% cross-sport blend. Only frozen Core-eligible NFL entities receive score credit. Valid outside-Core selections remain evidence.
Temporal evidence needs review
Provider evidence, publication dates, retrieval timestamps and explanations are retained. Objective in-window dates can support automatic acceptance. Ambiguous or out-of-window evidence requires human review.
A retrieval timestamp records when evidence was received. It does not prove that the underlying event happened within the observation window.
APPROVED DIAGNOSTIC COMPARISON
Relative gap = Recent cohort percentile − Established cohort percentile
Recent and Established raw scores remain separate because they measure different constructs. Tie-aware within-cohort ranks or percentiles can compare relative standing. A positive gap means stronger relative Recent prominence. It does not, by itself, prove upward movement. Movement needs multiple compatible Recent observations.
06
Freeze the instrument. Preserve the evidence.
Publication depends on completeness, identity resolution, reproducible scoring and an audit trail.
- 01
Collect
Retain the full visible answer with its prompt, model, dates and provider evidence.
- 02
Validate
Reject malformed structure and operational failures through bounded, documented retries.
- 03
Adjudicate
Resolve identity and eligibility while preserving original labels, ranks and answer positions.
- 04
Audit and publish
Require complete accepted slots, deterministic scores, review records and an approved publication snapshot.
Publication evidence
- Frozen prompts, providers and model identifiers
- Accepted responses and exact observation relationships
- Original ranked labels and positions
- Identity and eligibility decisions
- Deterministic scores and publication audit
LEGACY SERIES
Archived separately
Earlier Bowstone observations used a different prompt and scoring architecture.
Read the legacy methodologySEPTEMBER 26, 2026
Published Core begins
The first compatible Core observation establishes the current series boundary.
VERSIONED FUTURE
Prospective change
New versions receive their own contract and fingerprint. Historical evidence remains immutable.
07
Read every result with its limits.
Measurement discipline improves interpretability. It does not make model output equivalent to ground truth.
Probabilistic answers
The same question can produce different answers across repeated runs. Frozen wording and model settings make observations comparable, but they do not remove model variance.
Different provider behavior
Models differ in training, retrieval tools and ranking behavior. Search availability does not prove that retrieval occurred in a particular response.
Not a consumer survey
AI-answer visibility is not a representative survey of fans, consumers or the public. It measures model output under a defined protocol.
Recognition is not popularity
A model selecting an entity does not establish real-world popularity, favorability or objective brand quality.
Events still require reporting
Discovery preserves what models say, but it does not independently prove the events or explanations in those answers.
Zero needs context
A complete observation with no selection is a measured zero in that instrument. It is not proof that an entity has no recognition or real-world relevance.
Instruments are not interchangeable
Established, Recent and Discovery scores answer different questions. Raw scores should not be subtracted or combined without an approved comparison method.
No commercial prediction
Bowstone scores do not establish sponsorship value, franchise valuation, financial performance or future commercial outcomes.
MEASURED. REVIEWED. PUBLISHED WITH CONTEXT.
Social Presence measures comparable audience momentum.
BX combines that weekly momentum with Established AI Visibility for eligible teams and players.
Social Presence
Instagram and X follower growth across shared, usable platforms in two comparable weekly observations.
Weekly momentum, not audience sizeBX Overall
A type-pool composite for teams and players across supported sports. Teams compare with teams and players with players.
65% AI + 35% SocialBX peer view
The same component weights can be normalized within an entity's league or cohort for a peer comparison.
Compatible peer populationBX FOR ELIGIBLE TEAMS AND PLAYERS
The weights are fixed. Without qualifying social data, BX is unavailable. The system does not substitute an AI-only composite.
COMBINED WEEKLY SOCIAL GROWTH
(Current comparable total ÷ prior comparable total) − 1
Only platforms usable on both dates enter the ratio. A platform missing or flagged for review on either date is excluded from both sides, not filled with zero.
Comparable inputs first
One valid platform may qualify. Known handle changes, unresolved anomaly flags and noncomparable dates exclude affected inputs. The date window is 5 to 9 days across consecutive Saturday cycles, with the latest social observation no more than three days old at the benchmark date.
Without a qualifying comparison, Social Presence and BX remain unavailable. A measured zero growth rate is valid. Missing or noncomparable growth is null.
Normalize within the right pool
Overall Social Presence normalizes teams against eligible teams and players against eligible players. Peer variants use the same entity type within the cohort.
Social Presence = 100 × actual combined growth ÷ the largest positive eligible growth rate, bounded from 0 to 100. Each entity's contribution to the denominator is capped at 12% to limit a single extreme observation.