Expanding abroad: the SEO mistakes that cost you
TL;DR
- Translating your site isn't entering a market: people search differently and keywords don't translate literally.
- 76% of consumers prefer to buy in their own language and 40% won't buy in another - localization is revenue, not decoration.
- hreflang errors, local engines (Baidu/Yandex/Naver) and scaled machine translation (which Google treats as spam) sink expansions.
- You need someone who masters the language AND the market's SEO/GEO - not a translator who doesn't know local search.
The most expensive illusion in international expansion is thinking that "translating the site" means entering a market. It doesn't. People in each country search with different words, trust different signals and, in several markets, don't even use Google. A linguistically perfect translation can be commercially useless - and most translators don't do SEO or GEO.
Translating isn't expanding
The most-searched term in a market is rarely the literal translation of the original: intent shifts, volume shifts, synonyms shift. Slogans and value propositions need transcreation (cultural recreation), not word-for-word conversion. And the data is blunt: per CSA Research, 76% of consumers prefer to buy with information in their own language and 40% won't buy on sites in another language. Localize badly and you shut the door on a huge slice of the market.
Where Google isn't even in charge
Optimizing only for Google in markets ruled by other engines is optimizing for the wrong engine:
| Market | Dominant engine | Google's share |
|---|---|---|
| China | Baidu (~47%) | ~1.8% |
| Russia | Yandex (~71%) | ~27% |
| South Korea | Naver (contests the lead) | ~47% |
Each has its own rules - Baidu favours local hosting and domains (and ICP registration); Yandex and Naver have distinct signals and result formats.
Scaled machine translation is spam to Google
Running your whole site through machine translation, without review or added value, falls under Google's scaled content abuse policy - which explicitly lists translation among the ways of generating low-value content at scale. Google doesn't penalize machine translation itself; it penalizes value-less content, whatever the method. Mass-translating just to "have the site in 10 languages" is a risk, not a shortcut.
hreflang: one mistake breaks it all
hreflang tells engines which version to serve each user. It's powerful and fragile: miss a return tag or the self-reference and Google ignores the whole cluster. In a study of 20,000 multilingual sites, Semrush found hreflang conflicts in 58% of them. Without it done right, versions compete with each other (cross-language cannibalization) and dilute signals.
The angle almost nobody sees: GEO per language
AI engines answer in the user's language and weigh local relevance. A brand cited only in English can be invisible for the same question asked in German - simply because, in the German corpus, that brand doesn't exist as a credible entity. Getting cited by AI in a market takes content and entity signals in that language. It's the same GEO work, done market by market.
Why you need a specialist - not just a translator
A translator optimizes for linguistic fidelity. Expansion needs to optimize for demand, trust and conversion in the target market. That's a different skill. I ran the international sites of Mercedes-Benz as Global SEO Specialist and I'm bilingual PT/EN: I bring language, technical SEO (hreflang at scale), local entity strategy and GEO in one person. If you're expanding, let's talk - see also international SEO and what working with me looks like.
Key data
| Prefer to buy in their own language | 76% |
|---|---|
| Won't buy in another language | 40% |
| Sites with English content | ~49% |
| Google in China (Baidu 47%) | 1.76% |
| Google in Russia (Yandex 71%) | 27% |
| Multilingual sites with hreflang errors | 58% |
CSA data from 2020 and the Semrush hreflang study from 2017; per-market shares (China) are volatile — 2026 values.
Sources: CSA Research — Can't Read, Won't Buy · W3Techs — Content languages · StatCounter · Semrush — hreflang mistakes
Common mistakes and how to do it right
I've seen brands burn expansion budgets by treating international SEO as a translation job. The costliest mistakes:
| Avoid | Do |
|---|---|
| ✗ Translating keywords literally | ✓ Do native keyword research per market - intent and volume shift |
| ✗ Mass machine-translating the site without review | ✓ Localize with real value - Google treats scaled MT as spam |
| ✗ Leaving an hreflang cluster without self-reference | ✓ Validate hreflang, return tags and ISO codes at scale |
| ✗ Optimizing only for Google where Baidu or Yandex rule | ✓ Optimize for the engine that dominates the target market |
| ✗ Assuming the brand is recognized everywhere | ✓ Build local entity and signals (address, author, mentions) in the target language |
Tools I use and recommend
- Google Search ConsoleOrganic performance, coverage and field Core Web Vitals.
- Screaming Frog SEO SpiderDesktop crawler for deep technical and on-page audits.
- SemrushKeywords, site audit and competitor analysis.
- GeminiGoogle's AI assistant, grounded in Search.
- PerplexityAnswer engine with sources, useful for research and GEO.
Sources
- CSA Research - Can't Read, Won't Buy
- Google - Managing Multi-Regional and Multilingual Sites
- Google - Spam Policies (scaled content abuse)
- StatCounter - Search Engine Market Share
- W3Techs - Content Languages
- Semrush - Common Hreflang Mistakes
Frequently asked questions
Isn't translating my site enough?
No. Translation is linguistic; expansion is finding demand, earning trust and converting in the target market. That takes native keyword research, hreflang, local entity signals and, now, GEO per language. GEO guide →
ccTLD, subdomain or subfolder?
It depends on resources and ambition: a ccTLD (.de) sends a strong country signal but forces you to build authority on each domain; a subfolder (/de/) consolidates authority and is easier to maintain. Google doesn't infer country from the URL alone - it needs hreflang and configuration. Google official docs →
Does machine translation hurt SEO?
At scale and without value, Google treats it as spam (scaled content abuse). As a starting point reviewed by someone who knows the market, it's fine; published en masse without review, it's a risk. E-E-A-T and entities →