Why Overseas Buyers Can't Find You in AI

Many export businesses hit something they can't explain: a website written up over more than a year with decent Chinese search rankings, yet overseas buyers — asking AI in English which Chinese supplier in a given category is any good — never see them in the answer.

The conclusion first: usually it isn't weak optimization — it's a language mismatch. Buyers ask in English, and retrieval looks for pages in the language of the query; if all you have is Chinese content, you were never in the candidate pool to begin with.

One boundary: this doesn't mean Chinese content is useless, nor that you must go fully multilingual right away. The key is which language your customers actually ask in. Below is the mechanism first, then what to do — and which route doesn't work.

The official rule: retrieval looks for pages in the query's language

No guessing needed — Google's official documentation states it plainly.

In the official document "Managing Multi-Regional and Multilingual Sites", one line reads: "Google Search tries to find pages that match the language of the searcher." The document also notes that multilingual sites are distinguished as "multilingual" versus "multi-regional", and the matching logic differs between the two.

That explains the earlier puzzle: your page is fine in itself, but it doesn't match the query's language, so it gets filtered out at the retrieval step.

Scope note: this is Google's official explanation of Search mechanics, at the search-feature level. AI features (AI Overviews, AI Mode) sit on the same retrieval system, but their specific ranking logic isn't fully public — so this article states only the "language matching" rule and doesn't extrapolate to AI's full ranking mechanism.

Language is judged by visible body text — not the lang attribute, not the URL

This is the point most often got wrong — and why many people configure multilingual settings and see no effect.

The same official document has another line: "Google uses the visible content of your page to determine its language. We don't use any code-level language information such as lang attributes, or the URL."

What that means in practice.

• Changing a Chinese page's lang="en" doesn't make it an English page;
• Changing the URL to /en/xxx.html doesn't make it an English page either;
• What actually counts is the words visible on the page.

So "adding a multilingual tag to a Chinese page" doesn't fix a language mismatch. The only fix is to genuinely write English content.

Machine translation: the official guidance says it's a rough road

So can you batch-generate English pages with a translation tool? The official documentation answers this too.

The same document makes two points: first, translating only the template text (navigation, footer) while the body stays in one language degrades the user experience; second, more directly — automated translations "may not make sense" and can be treated as spam; more importantly, a bad or stilted translation harms how your site is perceived. The document even suggests using robots rules to stop search engines from crawling auto-translated pages on your site.

This matters: machine translation isn't just "less effective" — it can be judged as spam and drag down the credibility of the whole site. For a company doing GEO (Generative Engine Optimization), that's worse than not doing it at all, because GEO rests on whether AI is willing to trust you.

The right approach: separate URLs + two-way annotation + genuinely writing in the target language

The official path isn't complicated — three things.

• A separate URL for each language. Google recommends a different URL per language version, rather than switching dynamically via cookies or browser settings.
• Annotate language variants in both directions. Use hreflang to tell search engines "these pages are localised versions of the same content". Google's hard rules: each language version must list itself and every other version; annotations must point both ways, since one-way tags are ignored; and URLs must be fully qualified (including http/https).
• The body text must genuinely be written in the target language. Language is judged by visible body text — which takes us back to the second section.

Google also recommends setting a fallback page for unmatched languages (marked with x-default) when no version exists for a given language, rather than dumping users into some arbitrary language version.

One easy trap: don't auto-redirect by language

Many sites do this: detect the visitor's country and jump them to the matching language version. It sounds considerate, but it's unfriendly to search engines.

The documentation explains why: Googlebot usually crawls from the US and sends requests without an Accept-Language header. The result — sites that auto-redirect based on language settings may not have all their language versions fully crawled.

The right approach: let users switch language themselves via clear links, rather than auto-redirecting. Google also advises against redirects based on guessing the user's language.

Three common questions

My customers are mostly overseas — was the Chinese content wasted?

No. It depends where your customers ask. If they're overseas buyers asking in English, Chinese content genuinely doesn't connect; but if domestic buyers or channel partners also ask, Chinese content still has value. The practical move: keep producing Chinese content and add English as a separate line — not as a replacement.

Do we need every language?

No. Start with the one your customers ask in. If they cluster in English-speaking markets, do English first; if one market dominates, consider that language. One complete, properly cross-annotated language version beats five half-finished ones — because language variants only work when they point both ways.

What should the English content cover?

Same principle as Chinese content: write the questions customers actually ask, not a product manual. "How do I choose this spec", "what certification is needed to export to this region", "how are minimum order and lead time negotiated" — people ask these in English too, and the competition is often thinner than in Chinese.

In closing

To sum up: when overseas buyers can't find you, it's usually not that the content is poor — it's that the language doesn't match.

The order: confirm which language your customers ask in → write the body in that language → host it on separate URLs → annotate with two-way hreflang → don't auto-redirect. The time machine translation saves comes back doubled in lost credibility.

One small thing to start with: take your three most important questions, put them to AI in the language of the market you most want, and look at which sites get cited and what language they're in. That alone tells you whether the problem is "nobody's asking" or "the language doesn't match".

Yihang Network, based in Shenzhen, previously worked in cross-border B2B and Alibaba.com-related services, and expanded into GEO from 2024. We're doing the same thing on our own website — building Chinese and English as two separate lines, because content has to match wherever the customer is asking.

Data and scope note: The retrieval mechanics, language-determination rules, guidance on translated content, hreflang rules and crawling notes in this article all come from Google Search Central documentation ("Managing Multi-Regional and Multilingual Sites", "Localized Versions of your Pages") and Google's official help documentation. Official rules describe how a feature works; they do not represent any specific company's results. The methods listed here raise the probability of being retrieved and cited; they do not guarantee a company will appear in AI answers, and no timeline or business result is promised.

1. Google Search Central, "Managing Multi-Regional and Multilingual Sites"
2. Google Search Central, "Localized Versions of your Pages" (hreflang and x-default rules)
3. Google Search Central, spam policies (including guidance on automated translations)
4. Google Help, "AI Overviews in Google Search" (available regions and languages)
5. Google official blog, AI Overviews expanded to 100-plus countries with multilingual support (2024-10)