Glossary
Fourteen terms.
Action (HOLD / HEDGE / SEO_PLAY / ARBITRAGE)
The recommended next move for a category based on the gap between AI vocabulary and market evidence. HOLD = AI and evidence agree, maintain positioning. HEDGE = signal is unstable, monitor before investing heavily. SEO_PLAY = buyer search language differs from AI language, build content bridges. ARBITRAGE = AI/market mismatch may create an exploitable visibility gap.
Anchoring
Which data source AI's vocabulary for a category is closest to. "Anchored to OEM" means AI describes the category the way manufacturers do. "Anchored to search" means AI learned the category from buyer search behavior. The anchoring signal tells you whose language AI absorbed.
Brand category peers
Other brands that compete in the same category. Category-level peers, NOT SKU-level drop-in replacements. Sourcing-desk starting point for "who else operates here?" — cross-reference against your own inventory and supplier relationships before acting on the list.
CategoryRank Ontology
The continuously-refreshed, OEM-anchored category graph underneath every verdict. 253 total categories (174 with consistent AI coverage this week). We optimize the ontology to maximize semantic closeness to OEM canonical names while amplifying coverage through distributor and aggregator vocabularies; new categories integrate as they emerge in the wild, and drift is tracked week-over-week instead of smoothed into a single snapshot.
Centroid
The mathematical center point of all AI observations for a category, represented in 1024-dimensional space. Think of it as the "average" of everything AI says about a category. We compare each category's AI centroid against OEM, distributor, aggregator, and buyer-search centroids to measure vocabulary distance.
Channel pull
Per-(brand, AI cohort) attribution of which external data source AI's category language is closest to: OEM-aligned, distributor-aligned, aggregator-aligned, or canonical-aligned. Says "when AI describes this brand's categories, whose vocabulary is filling the vacuum."
Cohort-stable
A category where AI brand coverage is consistent across our panel of 16+ frontier LLMs. Cohort-stable categories have reliable verdicts. Categories that are NOT cohort-stable have verdicts flagged as partial or insufficient.
Convergence score
For a (brand x category) cell, the share of observations using the dominant alias form. 100% = the funnel agrees. <50% = fragmented. Computed per cohort (US-AI, EU-AI, CN-AI, Other-AI) so geographic disagreement surfaces.
Lens
Your operator perspective. CategoryRank shows the same category data through four lenses — OEM, distributor, aggregator, and broker — because each operator type needs different actions from the same underlying signal. Same substrate, different operating brain.
Showcase tier
Select CategoryRank Ontology categories unlocked publicly as marketing surface (currently power-supplies; others added as we onboard demo customers). Anonymous visitors and crawlers can read full data on showcase categories. All other categories show framework + counts only.
Substrate
The complete time-indexed evidence base behind every CategoryRank verdict. Currently 21 weeks of weekly observations (since 2025-W51), 26M+ AI observations, 35,000+ brands tracked across 16+ frontier LLMs. The substrate is the moat — the methodology is reproducible, the accumulated time-indexed data is not.
Top-form share
For a (data source x category) cell, the count of the most-frequent alias form divided by total observations. The fragmentation signal. Low share = vocabulary fractured. High share = vocabulary converged.
Vantage (data source)
One of four perspectives we measure category language from: AI (LLM-emitted forms), OEM (brand product-page taxonomy), distributor (parametric trees from authorized resellers — Mouser, DigiKey, Arrow, Avnet, LCSC), aggregator (cross-distributor discovery surfaces — Octopart, Findchips, Oemsecrets). Same category; four sets of vocabulary.
Vocabulary fragmentation
The structural disagreement between data sources on what to call a category. CategoryRank's core measurement. Where it's high, sourcing buyers find arbitrage; where it's low, the market is efficient and brand competition is on parametrics, not naming.