Best Western France:Does ChatGPT know the network?

French + FR-proxy is the primary pass, not English + US

Most of this site’s cross-vertical AI-search field tests default to English-only, US-proxy-only prompts, matching an international-traveler assumption. That convention doesn’t transfer to a brand study of Best Western France specifically, because most of its clientele isn’t that. The evidence comes directly from Best Western France’s own 2025/2026 press releases (pressroom.bestwestern.fr):

National context backs this up directionally: INSEE puts France’s 3-star hotel clientele at roughly 40-47% foreign depending on season — but that figure is itself Paris/Riviera-weighted at the national level, consistent with the other ~85% of this network (205 of 242 reconciled cities, everywhere outside Paris+PACA) sitting well below it, toward domestic French.

Design consequence: two passes, not a uniform cross-product. A primary pass (French, FR-proxy) across all 242 reconciled cities, matching the real, majority-domestic, majority-business clientele across the network; and an add-on pass (English, US-proxy) scoped to Paris + all 36 PACA cities (37 total) — exactly where Best Western’s own reporting says foreign leisure clientele concentrates. 8,370 distinct prompts, every one unbranded (no “Best Western” anywhere in the text): the question is whether the brand surfaces when someone asks a generic hotel question about a destination where it has a property, not whether the AI can describe a hotel the user already named.

Star/budget tier and persona: where the brand nearly disappears

Best Western France’s real inventory clusters at 3-4 star. “Luxury hotel in X” scoring higher than “3-star” is a real, if mildly counterintuitive, AI-behavior finding: a small/mid-size town often has no genuine luxury option, and the AI falls back to the most prominent hotel it has — frequently the local Best Western. “Cheap hotel” queries, by contrast, pull in budget chains (Ibis Budget, Premiere Classe, F1) that undercut Best Western on price positioning even in towns where it’s the only real full-service option.

Persona: business travelers see it far more than couples

Consistent with the methodology section’s finding from Best Western France’s own press data (Corporate revenue 15x Leisure) — the AI models’ associative sense of the brand skews business/roadside too, not just the real booking mix.

Soft-brand recognition: the flagship name dominates

85% of the time an AI answer names a Best Western property, it uses a form close enough to the flagship “Best Western [Plus/Premier/Signature] …” naming to match exactly. Soft-brand properties (Sure Hotel, BW Premier/Signature Collection) are recognized far less reliably — even after fixing the resolver’s brand-token list, ~12% of hits still needed the fuzzy/alias/review-flagged paths rather than an exact match.

Who shares the answer

A mention rate tells you how often a brand appears. It cannot tell you who appears next to it — and on an unbranded panel, whatever else is in the answer is, by construction, the set the model considered for the same query. So we counted it: for every chain, how often it shares an answer with Best Western.

Two numbers, both from one joint count, and both bounded by 100% because the unit is the answer rather than the mention — a brand named three times in one answer counts once. Share of Best Western’s answers is P(chain | Best Western). Share of its own answers is the reverse, P(Best Western | chain). They are deliberately different: Campanile appears in 6.8% of Best Western’s answers, but Best Western appears in 59.7% of Campanile’s. That asymmetry is the size difference between the two brands, not a contradiction.

Lift is the column that matters. It divides P(chain | Best Western) by the chain’s own base rate across the whole panel: 1.00 means the two brands share answers exactly as often as chance would produce, given how often each appears at all. Without it a co-occurrence table is just a popularity ranking wearing a competitor label.

The faint structure that does exist runs along positioning, and it runs both ways. Slightly above chance sit the French mid-market and budget ladders — Campanile 1.14, Holiday Inn 1.14, The Originals 1.13, Première Classe 1.13, Kyriad 1.10, Brit Hotel 1.09. Slightly below sit design, upscale and international brands — Mama Shelter 0.18, Marriott 0.31, Adagio 0.70, Hilton 0.73, Radisson 0.75. Best Western is not merely absent from upscale answers; it is displaced from them, which is the same story the Honfleur and Lyon zero-mention cities tell above, measured across the whole panel instead of anecdotally.

Language moves it. On the English/US-proxy add-on, the international brands climb next to Best Western while the French budget ladder falls away: Holiday Inn goes 1.04 → 2.06, The Originals 1.06 → 2.11, Hilton 0.72 → 1.09, while Première Classe drops 1.09 → 0.50. Read those English figures as directional only — the add-on is 2,911 captures against 19,770 French ones, and several chains sit on double-digit sample sizes. The full split, sample sizes included, is in the dataset below.

One caveat on the denominator: this section identifies Best Western by name match on the extracted hotel list, which reaches 52.6% of these answers. The 64.8% headline elsewhere in this study comes from the full resolver, including the fuzzy and alias tiers. The gap is the soft-brand recognition problem measured in the previous section, so treat the co-occurrence rates as a floor.