Cross-border brands: splitting GEO between China and abroad

This guide is for brands that sell both in China and abroad, and for the team leads who have to split the GEO work. It covers how to divide platforms, sites, content, third-party channels and people so the two markets don’t undercut each other. Setting up the site structure takes a few weeks. The content and coverage work runs continuously in both markets. The platform-specific detail lives in two companion guides: GEO in China: Doubao, Yuanbao, Qwen, ERNIE, DeepSeek and Kimi and GEO outside China: Google AI Overviews, ChatGPT, Perplexity and Copilot.

The short version

1. Treat China and abroad as two programmes that share one set of facts.
2. Give each language its own URL and link the versions with hreflang.
3. Open each market’s crawlers: Baidu and the Chinese engines at home, Google, Bing, OpenAI and Perplexity abroad.
4. Write natively for each market instead of translating one source text.
5. Build third-party presence where each market’s assistants actually read.
6. Split the team by market, with one owner for the shared facts.
7. Measure each market separately, in its own tools and apps.

Source preferences in China are observations from industry studies and change over time.

1. Two programmes, one set of facts

The AI platforms in the two markets are different products with different sources. Even inside China, platforms share few sources.

Observed in 2025-10, more than 9 months ago. Answer engines change often; re-test before relying on this.

Growthbox and Percent Technology monitored seven Chinese AI platforms from July to October 2025. Of 51,000 sources cited, only 9.34% were cited by four or more platforms.

Observed in practice · observed 2025-10 · 增长黑盒 · 百分点科技 (unpublished report, not publicly available; readers can't check it)

Meltwater’s 2026 report on Chinese brands going abroad argues that content built up on WeChat, Weibo, Zhihu and Bilibili is hardly crawled or cited by mainstream English-language models. It also argues that machine-translated content won’t pass AI selection. The report states this as its view and shows no test of it.

Hypothesis · Meltwater 融文 (unpublished report, not publicly available; readers can't check it)

So the work in China does little for answers abroad, and the reverse. What the two programmes share is the facts: product names, specs, prices, certifications and company details. Keep one master list and make both markets draw from it.

Two markets, two playbooks, one fact sheet. Most cross-border mistakes come from letting the facts drift apart.

2. One URL per language, linked with hreflang

Search engines and AI systems need a stable address for each language version.

“Google recommends using different URLs for each language version of a page rather than using cookies or browser settings.” Google also advises against automatically redirecting users to another language version, and says it determines a page’s language from its visible content.

Officially documented · Google Search Central

If two pages don’t both point to each other, hreflang tags will be ignored. Each language version should list itself as well as all other versions.

Officially documented · Google Search Central

• Use subfolders, subdomains or separate domains. Pick one and keep it (URL structure).
• No automatic language or IP redirects. Offer a language switcher instead.
• Only add hreflang for real translations (Set up hreflang for translated pages).
• Don’t canonicalise translations to the source page (Canonical URL signals).
• Hosting for mainland China has its own licensing and network questions. They sit outside this guide.

3. Open each market’s crawlers

Each market has its own gatekeepers. A single robots.txt or firewall rule can quietly close one market while serving the other.

Baiduspider follows robots.txt, and blocking Baiduspider removes a site from Baidu search and other search results Baidu provides.

Officially documented · Baidu

OpenAI: “Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links.”

Officially documented · OpenAI

Market	Crawlers to allow	Detail
China	Baiduspider, 360Spider, Sogou, Bytespider	GEO in China: Doubao, Yuanbao, Qwen, ERNIE, DeepSeek and Kimi
Abroad	Googlebot, Bingbot, OAI-SearchBot, PerplexityBot	GEO outside China: Google AI Overviews, ChatGPT, Perplexity and Copilot

Firewalls are often tuned to one market’s traffic. Check that the other market’s crawlers aren’t challenged or blocked (AI search access audit, Verify AI crawler traffic in server logs).

4. Write natively for each market

Each market’s buyers ask different questions and trust different proof. Translate the facts, not the article.

Google warns that translating only the boilerplate text of a page while keeping most of the content in a single language can create a poor user experience.

Officially documented · Google Search Central

• Build each market’s prompt list from its own customers’ questions.
• Have a native writer in each market draft from the shared fact sheet.
• Use local units, currencies, standards and certifications.
• Mass-produced pages in many languages can fall under Google’s scaled content policy (Scaled content abuse).
• Writing for citation: For content editors: writing articles AI answers can quote.

5. Third-party presence where each market reads

The sources assistants cite differ sharply between markets. Abroad, reviews and community sites carry weight. In China, each assistant leans on its own group’s platforms.

Meltwater says its analysis of more than 8 million AI citations found reviews and social proof made up 57% of citations, and white papers and other thought-leadership content 5.4%. The report doesn’t say which assistants, markets or period were sampled.

Observed in practice · observed 2026-07 · Meltwater 融文 (unpublished report, not publicly available; readers can't check it)

In a Meltwater case study of open-ended phone-buying questions to AI models, brand websites were the most cited source. YouTube came second with 1,300 citations, and Reddit had 542. Tech media such as Tom’s Guide, TechRadar and CNET dominated specification comparisons. The report doesn’t disclose the number of prompts or the sampling dates.

Observed in practice · observed 2026-07 · Meltwater 融文 (unpublished report, not publicly available; readers can't check it)

In China, a pharma study by IQVIA and a partner found Doubao leaning to ByteDance’s Toutiao and Douyin, Yuanbao to WeChat official accounts and Tencent News, and Wenxin to Baijiahao.

Observed in practice · observed 2026-06 · 艾昆纬（IQVIA）· 上海零假设信息科技 (unpublished report, not publicly available; readers can't check it)

	China	Abroad
Owned channels	WeChat official account, Toutiao, Baijiahao, Zhihu	Own site, YouTube channel
Third-party	Dianping, Baike, vertical media, complaint sites	Review sites, Reddit, trade and tech media
Map it with	Which sources Chinese AI assistants cite	Google AI features vs ChatGPT search vs Perplexity

Coverage built in Chinese media rarely helps abroad. Plan English-language (or local-language) coverage as its own budget line (PR and brand teams: how press, encyclopedias, Q&A sites and reviews shape AI answers).

Your reputation doesn’t travel with your product. Each market’s assistants only know what that market’s sources say about you.

6. Split the team by market

This is a working recommendation, not a documented rule.

Meltwater recommends a central team plus regional teams, rolled out in phases: fix the basic brand facts in the first three months, build deeper content in months three to six, then run ongoing monitoring. The report presents this as advice without evidence of results.

Hypothesis · Meltwater 融文 (unpublished report, not publicly available; readers can't check it)

A split that follows from the steps above:

Role	Owns
Central (one owner)	The fact sheet, site structure, hreflang, crawler access, rules on claims
China team	Chinese site, ecosystem accounts, Chinese media, China measurement
Overseas team(s)	Local-language sites, review and community presence, overseas media and measurement

Both teams must follow both markets’ rules. They differ in detail.

The US FTC rule makes it a violation “for a business to write, create, or sell a consumer review, consumer testimonial, or celebrity testimonial that materially misrepresents” that the reviewer exists or used the product. It also covers buying reviews and fake social media indicators.

Officially documented · US Federal Trade Commission (eCFR)

China’s 2026 Qinglang campaign on AI misuse targets AI data poisoning, including using GEO for malicious marketing, and fake-account networks that mass-produce near-identical posts.

Officially documented · 新华网（中央网信办通知）

Background: GEO regulation in China, Google spam policies.

7. Measure each market separately

Each market needs its own tools, apps and prompt sets.

Search Console’s Generative AI performance report covers AI Overviews and AI Mode and can be grouped by country.

Officially documented · Google Search Console Help

A 2026 Hefei study found Chinese assistants’ APIs and consumer apps could give opposite results: DeepSeek recommended a studio in 69% of API runs but 1% of app runs.

Observed in practice · observed 2026-08 · 腾讯新闻（报道合肥图灵象普拉提行业GEO白皮书）

• Abroad: Search Console by country, Bing AI Performance, sampled prompts in ChatGPT and Perplexity.
• China: sampled prompts in the consumer apps, on phones, in the target city.
• Keep one dashboard with the two markets side by side, never merged into one score (For marketing leads: measuring GEO and reporting it upward, Measuring AI visibility by sampling prompts).

What not to do

• Don’t auto-redirect by IP or browser language. Crawlers may never see the other versions (step 2).
• Don’t machine-translate the Chinese site and call it done. Google judges language from visible content, and buyers read it too (steps 2 and 4).
• Don’t assume Chinese media coverage reaches overseas assistants. The source pools barely overlap (steps 1 and 5).
• Don’t let prices or specs differ between language versions. Assistants repeat whichever they find (When AI gets your brand wrong: find it, fix it, prevent it).
• Don’t buy reviews or mass-post “mentions” in either market. Both the FTC rule and the Qinglang campaign cover it (step 6).

How you’ll know it’s working

• Each language version is indexed in its target engines and linked by valid hreflang.
• Verified crawler visits from both markets’ bots in your logs.
• Per-market mention rates rise on the same prompt sets over months.
• Facts in AI answers match the fact sheet in both languages.

Evidence at a glance

Recommendation	Evidence
Separate programmes per market	Observed (China), hypothesis (cross-market)
One URL per language, no auto-redirect, hreflang	Official
Open each market’s crawlers	Official
Native writing instead of translation	Hypothesis
Third-party presence per market	Observed
Central plus regional team	Hypothesis
Follow both markets’ rules	Official
Measure each market separately	Official and observed