Luxury brand AI visibility in China — by tier.

Why luxury is structurally different

Most generic GEO advice — "build SMZDM presence, get on Xiaohongshu, post on Bilibili" — is correct for aspirational luxury (handbags ≤¥10,000, watches ≤¥30,000, premium-but-not-prestige fashion). It is wrong above that price tier. Mainland Chinese consumers and decision-makers researching ultra-luxury reach for different surfaces, and the AI engines reflect this.

On our 1,620-response handbag panel (May 2026), DeepSeek's SMZDM mention rate was 100% in the aspirational tier (≤¥10,000), 71% in the mid-luxury tier (¥10K–30K), and 33% in the ultra-luxury tier (¥30,000+). The same brand at different price tiers gets surfaced via different source mixes — optimisation is not symmetric across the portfolio.

For details on the SMZDM tier collapse and the replacement source stack, see our SMZDM AI visibility analysis.

How DeepSeek, Qwen and Doubao surface luxury

Source-graph priority by luxury tier

Aspirational tier (≤¥10K handbags / ≤¥30K watches / premium beauty)

• SMZDM — category placements, "好价" deals, editorial roundups.
• Xiaohongshu — KOC seeding (10K–500K followers), tag/topic-page strategy. Detail.
• Zhihu — long-form category Q&A; topical-expert account aging.
• Bilibili — long-form review / unboxing content; particularly strong for watches, fragrance, and skincare on DeepSeek.
• Wikipedia / Wikidata — entity hygiene; the strongest DeepSeek mention-rate predictor in our 5-niche probe.

Mid-luxury tier (¥10K–30K handbags / ¥30K–80K watches)

Hybrid stack: aspirational sources still surface but with reduced weight; trade press (Vogue Business, WWD) starts to appear; brand-heritage content (founder story, atelier coverage) carries more weight than at the aspirational tier.

Ultra-luxury tier (¥30,000+ handbags / ¥80,000+ watches / fine jewellery / haute horlogerie)

• The Purse Forum — long-form authentication and collector community.
• Vogue Business / WWD — trade-press editorial.
• Sotheby's / Christie's / Phillips auction archives — provenance, lot history, comparable-sale context.
• Mainland-CN auction archives — 永乐 (Yongle), 中国嘉德 (China Guardian).
• Brand-heritage publications — collector magazines, foundation publications, museum coverage.
• Baghunter — niche but consistently cited for handbag-tier authentication and resale.

What changes a luxury brand can make

1. Tier-stratify your source-graph investment. Do not run a single Mainland-CN content programme across the entire portfolio. Map the tier of each SKU line and invest in the source stack each tier surfaces on.
2. Wikipedia / Wikidata entity hygiene first. Strongest single DeepSeek-mention-rate predictor in our 5-niche probe. Cleanly-cited Wikipedia entries with stable secondary sources surface across all tiers.
3. For aspirational tier: KOC seeding on Xiaohongshu + SMZDM editorial relationships + Zhihu category presence. Multi-quarter timeline; account aging matters.
4. For ultra-luxury tier: The Purse Forum participation (for handbags), Vogue Business / WWD trade-press relationships, auction-house provenance documentation, brand-heritage publication seeding. These are PR / editorial relationships, not paid placement.
5. Don't skip vertical specialists. Watch Home (腕表之家) for watches, premium-fashion vertical media — engines weight category specialists higher than generic luxury press.

What to avoid in luxury China AI visibility work

• Do not generalise the SMZDM 72% number. Handbag-panel-specific, tier-specific. Ultra-luxury collapses to 33%. Don't run an SMZDM-first programme for Hermès / Chanel exotic / Patek tier.
• Do not assume "more Xiaohongshu always wins". Xiaohongshu's weight collapses at ultra-luxury; The Purse Forum and Vogue Business take its place for handbags above ¥30K.
• Do not rely on brand-account posting alone. KOC content dominates citation share over brand-account posts on every Mainland source platform we measure.
• Do not promise "handbag rank in DeepSeek in 7 days." Mainland source-graph publishing compounds over quarters, not days. Account aging on Xiaohongshu / Zhihu alone takes weeks before AI citation rates rise.

How the Mainland-CN luxury research journey differs

The Mainland-CN luxury consumer journey runs on a different chronology than the Western journey, and the AI engines reflect this. Western luxury research is often a single, condensed editorial-and-influencer arc. Mainland luxury research is a multi-month process where AI assistants are now consulted at three distinct points — not one. The implication for source-graph investment is that any single-point optimisation strategy under-represents the brand on at least two of the three.

Stage one is category formation — six to twelve months before purchase, the consumer is researching the category itself (Why mechanical watches? Why exotic leathers? Why one Maison's heritage over another?). At this stage, Bilibili long-form videos, Zhihu category Q&A, and brand-heritage publications dominate the AI surface. SMZDM and Xiaohongshu come later. Brands absent from Bilibili long-form review and Zhihu category-question coverage at this stage are not in the consideration set the engines build later.

Stage two is shortlist construction — three to six months before purchase, the consumer narrows to a small handful of brands. Xiaohongshu KOC content, Vogue Business / WWD coverage and (for handbags) The Purse Forum threads dominate. SMZDM appears at the aspirational tier here; absent at ultra-luxury. Stage three is final selection — closer to purchase, where authentication, aftermarket value and provenance matter most. Auction-house archives, Baghunter and brand-heritage publications carry disproportionate weight. Eastbound's paid luxury audit runs all three stages as separate prompt panels because the source mixes diverge.

What an un-pivotable luxury brand looks like in AI answers

The brands that resist competitive substitution in AI output share three structural properties — properties luxury brands can actually build, not just claim. Asking "will Chinese AI swap your brand for a competitor?" is functionally a brand-defence question for any luxury client.

• Distinctive heritage with named-entity density — founder, atelier, archive, a date. Generic "since 1850" copy does not absorb; named ateliers, named master craftspeople, named heritage-piece references do.
• Category-defining language ownership — Hermès owns "Birkin"; Patek Philippe owns "perpetual calendar"; Cartier owns "Tank". When AI engines answer category questions, named-language ownership functions as an entity anchor that resists substitution.
• Multi-tier source-graph depth — coverage that spans editorial (Vogue Business / WWD), community (The Purse Forum, Watchuseek for watches), encyclopedic (Wikipedia, Baidu Baike), and provenance (auction archives). Brands present in only one or two of these layers are easier to substitute under prompt pressure than brands present across all four.

For the deeper measurement methodology behind these three properties, see the research piece on un-pivotable brands. For the action playbook on building each layer, see how to get cited on Chinese AI.