Geography and Query Alignment as Predictors of AI Brand Visibility: Evidence from a 13-Brand Single-Category Footwear Audit

Three prior papers in this series proposed a mechanism (Mention Density Model), supplied a measurementinstrument (MFBRA), and reported multi-category findings from a mixed-protocol dataset. This paper appliesthe same instrument under a uniform protocol to a single category - footwear - with 13 brands stratified bycountry of origin: US (5 brands), UK (4), Germany (2), Switzerland (1), and Spain (1). Each brand was auditedat full depth: 4 engines (ChatGPT, Claude, Gemini, Perplexity) x 10 queries each, producing 40 query-enginecells per brand and 520 cells total. This is the first paper in the series where every brand in the dataset ran thesame protocol - no Lean-tier (3q x 2m) or Balanced-tier (10q x 2m) brands. One exception: Birkenstock hadone Claude query error (39 usable cells instead of 40). Two patterns emerge. First, brands whose products align with the query set - running shoes, casual sneakers,everyday footwear - score substantially higher than brands whose products do not match, regardless of howwell-known the brand is offline. Hunter (premium Wellington boots, UK) returned SOV=0 across all fourengines across 20 combined queries in two independent audits; the brand is globally recognized but the queryset does not describe it. Birkenstock (sandals/clogs, Germany) scored 28; Dr. Martens (boots, UK) scored 20.The query-alignment signal is stronger than the geography signal in this dataset. Second, a raw origin-country gradient appears in the aggregate - US-origin brands average SOV 42.75, UKbrands 20 (27 excluding Hunter), German brands 16.5 - but this paper's central methodological argument is thatthe gradient does not survive as a country-of-origin effect. Every query-aligned brand in the sample is US-,Swiss-, or French-origin, so origin and alignment are collinear in the aggregate, and the raw gradient is whatthat collinearity produces. The two cells that isolate origin while holding alignment roughly constant - On(Switzerland, 58) and Hoka (France, 55), both query-aligned running brands - score above the US-originaligned average of 42.75, not below it. Two uncontrolled single-draw points cannot refute a hypothesis, butthey remove any support for one: there is no geography penalty where geography can be tested, and thedirection, where it can be read at all, runs the other way. Accordingly we separate two hypotheses the term"geography" conflates: H_country (origin label predicts SOV) and H_corpus (English-language editorialdensity predicts SOV). This dataset provides no independent support for H_country and its two alignmentcontrolled contrasts point against it, while remaining consistent with H_corpus: On and French-origin, USmarketed Hoka both built their recognition through dense English-language coverage, and both sit near the topof the distribution. This paper carries an important limitation that bounds even the alignment finding: it has no independentmeasure of corpus mention density. Every attribution of a score to "dense English editorial coverage" is posthoc, fit to the SOV it is meant to explain, and not falsifiable within this dataset. The corpus story is anchoredonly by a queued follow-up: English vs non-English mention counts for matched-product-type brands. Thequeued German brands (Adidas, Puma) test whether passport matters but not the Mention Density Model'ssharper native-language claim - they are English-corpus-saturated megabrands, not the thin-English-coverageEuropean brand that claim requires (Section 5.2). The finding is presented as a structured observation, not acausal claim. Keywords: generative engine optimization, GEO, AI search, brand recognition, share of voice, LLMevaluation, geography, mention density, footwear, single-category audit