Language pools are separate competitions

Engines answer German buyers from German-language sources, Japanese buyers from Japanese ones. Your English site's authority does not transfer — the question space, the trusted sources, and the corroborating third parties are all language-specific. This is the single most expensive lesson for outbound brands: English excellence plus machine-translated locale pages equals visibility in one market and absence in five.

Translation ≠ localization for AI

  • Question space differs. German buyers ask different questions (certification depth, DIN standards, warranty law) than US buyers (integrations, support SLAs). Translate the site, miss the matrix.
  • Source graph differs. Each market's directories, review platforms, trade press and forums are the corroboration layer — a German presence on German sources, not a .com/de page alone.
  • Trust markers differ. Local entity signals (local registration, local address, local phone, local case references) are what synthesis uses to decide you're a real option in that market.

The technical layer

  1. hreflang matrix: every page declares its language/region and its alternates, with x-default pointing to your default language. This site implements the pattern (en root, /zh/ subtree, x-default to en).
  2. URL strategy: subdirectories (example.com/de/) consolidate authority for most brands; ccTLDs only with real local operations behind them.
  3. Canonical honesty: self-referencing canonicals per language version; never canonical a locale page to the English original.

Measuring per market

Run the measurement protocol per market: localized prompt sets (buyer language, buyer locale), per-engine, and report per market. A global score is an average of unrelated competitions. Our China Outbound section applies this chapter market by market.

Frequently asked questions

Should I translate my whole site for GEO?
Translate the buyer-question coverage, not the sitemap. Twenty genuinely localized pages covering the market's question matrix outperform two hundred machine-translated ones — for engines and humans alike. Start with the clusters that drive decisions, in the market's own phrasing.
Does hreflang affect AI answers or just classic search?
Both. AI Overviews inherits Google's language/locale resolution, where hreflang is the declaration layer; conversational engines weight language match between query and sources. Getting the technical layer right doesn't win the competition — but getting it wrong forfeits it.