When this article was first written, artificial intelligence in Chinese luxury meant a virtual lipstick try-on and a chatbot that answered five scripted questions. That era is over. Today a Chinese shopper opens DeepSeek or Doubao, asks which brand of Swiss watch holds its value best, and reads the answer before ever touching a brand website. The question for luxury houses is no longer whether to add AI features. It is whether the machine mentions you at all, and what your sales staff do with the tools once the customer walks in.
I am Olivier Verot, founder of GMA. I have been in Shanghai since 2012 and my teams run WeChat CRM, Xiaohongshu and Baidu programs for European luxury and premium brands. I wrote this after eighteen months of watching clients deploy AI in China, some of it working, a lot of it quietly abandoned.
The Numbers That Actually Matter in 2026
Start with the market, not the technology. According to Bain & Company’s China personal luxury goods report published in early 2026, the Chinese personal luxury market contracted 3% to 5% in 2025, its second consecutive down year. Bain expects modest growth in 2026. That is the context in which every AI budget gets approved or refused.
The category split is more useful than the headline. Beauty grew 4% to 7%. Fashion fell 5% to 8%. Leather goods fell 8% to 11%. Watches fell 14% to 17%. Jewellery was roughly flat to down 5%, an improvement on 2024. If you sell watches in China right now, no AI tool fixes your problem. Your problem is pricing and resale value perception.
Two other Bain figures shape the AI conversation directly. Domestic purchases now represent 65% of Chinese luxury consumption, against 35% bought overseas. And the secondhand luxury market grew 15% to 20% in 2025, though it still sits below 10% of the primary market. Secondhand growth is the reason authentication technology stopped being a side project.
On the AI side, the scale is no longer debatable. CNNIC reported in its 57th survey, published February 2026, that China had 602 million generative AI users at the end of December 2025, a penetration rate of 42.8% and a jump of 141.7% over end of 2024. Out of 1.125 billion internet users. Your customer uses these tools daily. Your competitor’s marketing team does too.
What Digitalisation Did to Chinese Luxury Buying
A few years ago the reputation of a luxury brand in China depended almost entirely on face-to-face service in the boutique. Digitalisation and the weight of social platforms changed the way brands sell, promote and advertise. Chinese customers still want a distinctive experience in the store, but they expect the same standard online, and they now expect it inside WeChat rather than on your .com.
Daily life in China runs through apps: WeChat for messaging and payment, Xiaohongshu for discovery, Douyin for entertainment commerce, Tmall and JD for the transaction, Zhihu for the long answer. Every touch produces data. Companies collect behavioural signals at a volume that would be illegal or impossible in most of Europe. That asymmetry is the real advantage of running AI in the Chinese market, not the models themselves.
One thing worth correcting from the earlier version of this article: the idea that Chinese consumers do not care about data privacy is out of date. PIPL enforcement, consent pop-ups on every mini-program and public anger over facial recognition in shopping malls have changed the mood. Chinese shoppers are pragmatic about data, not indifferent. Collect what you need, tell them why, and give something back for it.
The Five Real AI Use Cases in Chinese Luxury Right Now
Here is what I actually see deployed on client accounts in 2026, and my honest read on each one.
| Use case | Where it runs | What it actually does | Verdict |
|---|---|---|---|
| Augmented sales advisor | Boutique tablet, WeChat Work | Pulls purchase history, sizing, gifting occasions and past conversations before the advisor replies | Highest return. Assists the human, never replaces them |
| Content personalisation by segment | WeChat CRM, mini-program | Serves different articles, invitations and product edits to VIC, aspirational and dormant segments | Works if your segments are real, useless if you have three tags |
| Authentication and counterfeit detection | Resale platforms, brand care service | Image and micro-detail comparison on stitching, serials, hardware | Now a trust requirement, not a differentiator |
| GEO, being the recommended brand | DeepSeek, Doubao, Yuanbao, Kimi | Structured, sourced Chinese content the models can read and cite | Most underinvested. Cheapest win available today |
| Localised content production | Xiaohongshu, Douyin, WeChat articles | Drafts, variants, translation groundwork, image adaptation | Useful for volume. Fatal if published without a Chinese editor |
1. The augmented advisor, not the replaced one
This is the one that pays. A luxury sales associate in a Shanghai boutique manages several hundred clients on WeChat Work. She cannot remember that a client mentioned her daughter’s wedding in March, that she takes a 38 in one line and a 39 in another, and that she returned a bag in 2024 because the strap was too short. The system remembers. It surfaces a briefing before the reply goes out.
The mechanism matters. The AI reads the conversation history and CRM record, drafts a suggestion, and the advisor sends it in her own words or ignores it. Nothing goes out automatically. Clients who let the machine send messages directly, and a few tried, got exactly what you would expect: polite, fluent, forgettable text that clients stopped opening. Response rates on WeChat one-to-one messages are the most honest metric in Chinese luxury. They collapse the moment the customer smells automation.
2. Personalisation by segment, which is not the same as personalisation
AI can adapt product suggestions using browsing behaviour, transaction history and offline interactions. That part of the old promise held up. What did not hold up is the fantasy of one-to-one content at scale. What works in practice is four to six well-built segments, each with its own content calendar inside your WeChat private domain: very important clients, repeat buyers, aspirational first-time buyers, dormant clients, gifting buyers, and often a separate tier-2 city group.
The gain is not creative, it is operational. A team that used to build one newsletter now builds one and lets the system assemble six versions from a component library. The VIC segment gets the atelier story and a private appointment link. The dormant segment gets a service reminder, not a product push. Same effort, roughly triple the open rate on the tail segments, based on what I see across client accounts.
3. Authentication and the secondhand problem
Secondhand luxury grew 15% to 20% in China in 2025. Chinese buyers increasingly treat handbags, watches and high jewellery as transferable assets, and platforms have built livestream inspection into the buying flow so the buyer sees serials, stitching and hardware in real time before paying.
Image recognition trained on micro-details now sits behind that flow. It flags suspect items faster than a human authenticator can, and the human confirms. For brands this cuts both ways. A strong resale market supports your primary prices, which is exactly what the watch category is missing right now. A weak one destroys them. The brands that engage with authentication, by publishing verifiable product references or offering a paid care and certification service, keep some control over how their brand gets priced. The ones that pretend resale does not exist end up with their pricing set by strangers on Xianyu.
4. GEO: being the brand the model recommends
This is the shift almost no luxury brand has budgeted for. With 602 million generative AI users in China, a meaningful share of high-intent research now happens inside a chat window. Someone asks Doubao for a gift under 20,000 RMB for a father who collects watches. The model returns three or four names. If you are not one of them, you were never in the consideration set.
Generative engine optimisation works differently from Baidu SEO. Chinese models pull heavily from sources they consider authoritative and structured: Baidu Baike, Zhihu answers, Xiaohongshu notes with real engagement, mainstream Chinese media, and increasingly WeChat official account articles. They favour content that states facts plainly, in Chinese, with attribution. A beautiful English brand film contributes nothing. A well-structured Chinese page explaining your materials, your workshops, your price positioning and your service terms contributes a lot. We covered the mechanics in detail after WeChat integrated DeepSeek, and the same logic now applies to Xiaohongshu used as a search engine.
5. Localised content production, with a hard rule
AI produces drafts, translation groundwork and image variants fast. On a Xiaohongshu account posting fifteen times a month, that is real time saved. The rule I give every client: no AI text reaches a Chinese audience without a native editor rewriting it. Chinese readers spot machine-written copy immediately. The tells are consistent, over-long sentences, mainland-inappropriate word choices, and that flat promotional register that reads like a state announcement.
There is also a legal dimension people keep missing. China’s Measures for Labeling AI-Generated Synthetic Content, issued by the CAC with MIIT, the Ministry of Public Security and NRTA, took effect on 1 September 2025. AI-generated text, images, audio, video and virtual assets distributed on Chinese platforms must carry both an explicit label visible to users and implicit machine-readable metadata or a watermark. Platforms share responsibility for unlabelled content. If your agency is generating Xiaohongshu visuals with AI, someone on your side needs to confirm the labelling is in place. Check it before your platform does.
Where AI Actively Hurts a Luxury Brand
Luxury sells on two things that automation is structurally bad at: relationship and scarcity.
A brand that automates its top-tier client relationship is shooting itself in the foot. The client who spends 500,000 RMB a year with you is paying, partly, for the fact that a specific person knows her. Replace that person with a bot and you have converted a luxury relationship into a customer service ticket. I have watched a European house do this to save headcount and lose a third of its VIC repeat rate inside two years. Nobody complained. They simply stopped answering.
Scarcity has the same problem. AI-generated content is infinite by nature. If your Xiaohongshu feed produces forty near-identical posts a month, you are teaching the algorithm and the customer that your brand is abundant. That is the opposite of what you sell. Fewer, better, slower is still the right instruction in this category. Use the tooling to raise the floor on quality, not to raise the volume.
And the old claim that AI is a differentiator no longer holds. Everyone has the same models. We wrote a whole piece on why AI stopped being an advantage in China. What remains proprietary is your customer data, your craft story and your people.
Case Study: Astrid, a Dutch Leather Goods House
Astrid runs a family leather goods house in the Netherlands, third generation, around 30 million euros of revenue, strong in Japan and Korea. She came to us in early 2025 with a Chinese WeChat account holding 14,000 followers and almost no commercial return.
The numbers were bad in a specific way. Her mini-program store converted at 0.4%. Average WeChat article open rate had fallen to 1.9%. Her Shanghai distributor’s two sales staff were handling around 600 client conversations each with nothing but a phone and memory. She was spending roughly 18,000 euros a month across content and paid traffic.
What she tried first, before us, was an AI chatbot on the mini-program and a batch of AI-translated product descriptions. The chatbot handled 70% of inbound messages and killed them. Sessions ending without a human reply went up, and the distributor reported clients complaining that the brand had become impersonal. The translated descriptions read like a manual. Neither thing was wrong technically. Both were wrong for the category.
What we changed took nine months. First, we removed the chatbot from client-facing conversations entirely and rebuilt it as an internal briefing tool inside WeChat Work: before answering, the sales assistant sees the client’s history, prior products viewed, and a suggested angle. Second, we split the 14,000 contacts into five segments and built a content library so each segment received different material. Third, we rewrote the whole Chinese product corpus with a native editor and structured it for machine reading: material sourcing, workshop location, care instructions, price bands, all in plain Chinese with sources, published on the WeChat official account and mirrored into Baidu Baike and Zhihu.
Why it worked: the AI stopped talking to customers and started helping the two people who could. And the structured Chinese corpus gave the models something to cite, which mattered more than we expected.
Results after nine months. Mini-program conversion moved from 0.4% to 1.1%. Open rate on the VIC segment reached 8.4%, while the dormant segment sat at 3.2%, both against a 1.9% flat baseline. Repeat purchase rate among identified clients rose from 12% to 19%. And on brand-relevant prompts we track in DeepSeek and Doubao, the house went from never appearing to being named in roughly a third of tests. Revenue in China roughly doubled off a small base, which is honest but worth stating: she started low.
What to Do in the Next Six Months
- Audit what the models say about you. Run twenty realistic Chinese purchase prompts through DeepSeek, Doubao and Yuanbao. Write down who gets recommended. That list is your real competitive set.
- Fix your Chinese corpus before buying tools. Structured, factual, sourced Chinese content on your own properties feeds both Baidu and the models. It costs less than any software licence.
- Point AI at your staff, not your clients. Briefing tools inside WeChat Work, never autonomous replies to a customer who spends real money.
- Check your AI content labelling. The rules have been enforceable since September 2025 and the platform carries liability with you.
- Segment properly, then personalise. Five real segments beat one fake algorithm.
- Decide your position on resale. Ignoring the secondhand market does not make it go away, it just means someone else prices your brand.
If you want the strategic and clientele side rather than the technology side, read our piece on luxury strategy to connect with the Chinese elite, and the profile of Chinese high net worth individuals.
FAQ
Should a luxury brand use an AI chatbot for customer service in China?
Not on the client-facing side, not for clients who buy. Use it for pre-sale logistics questions, opening hours, stock checks and order tracking. Anything touching product advice, gifting or an existing relationship goes to a human, assisted by AI in the background. Chinese luxury clients message brands on WeChat the way they message a friend, and they can tell within two replies whether they are talking to someone. Once they know, they stop.
What is GEO and does it replace Baidu SEO for luxury brands?
GEO means optimising to be cited by generative models such as DeepSeek, Doubao, Yuanbao and Kimi rather than ranked in a results page. It does not replace Baidu SEO, it sits on top of it, because the models pull from many of the same authoritative Chinese sources. In practice the work overlaps: clean structured Chinese content, presence on Baidu Baike and Zhihu, credible mainstream mentions, and Xiaohongshu notes with genuine engagement. The difference is you optimise for being quoted, not clicked.
Do we have to label AI-generated content in China?
Yes. The Measures for Labeling AI-Generated Synthetic Content took effect on 1 September 2025. AI-generated text, images, audio, video and virtual items distributed on Chinese platforms need a visible label for users plus machine-readable metadata or a watermark in the file. Both the content producer and the hosting platform carry responsibility. If you outsource content production in China, get written confirmation that your agency applies both layers.
Is AI worth the investment while the Chinese luxury market is shrinking?
The market fell 3% to 5% in 2025 and Bain expects only modest growth in 2026, so no one should be approving large platform budgets. But the two highest-return items here cost very little: a properly structured Chinese content corpus, and briefing tools for your existing sales staff. Both improve conversion on traffic you already pay for. Expensive AI showrooms and virtual try-on for handbags are the ones to postpone.
Work With Us on AI for Your China Luxury Strategy
GMA is a Shanghai agency and we run WeChat CRM, Xiaohongshu and Baidu programs for European luxury and premium houses. We build the Chinese content corpus the models cite, and the segmentation and sales-assist layer that turns WeChat contacts into repeat clients. We will tell you which AI projects to skip, which is usually most of them.
Contact us for a review of how Chinese AI assistants currently describe your brand.
We also work on Baidu SEO and PPC, WeChat development and CRM, Xiaohongshu and Douyin content, KOC and affiliate programs, PR and media exposure, e-commerce setup, and lead generation across Chinese platforms.
Some of our case studies
Over the years we have helped several luxury brands enter and grow on China’s digital platforms.
