International AI Visibility: One Brand, Many Languages, Many Engines

International AI visibility is not a translation problem. It is a consistency and sources problem, repeated per market.

Engines answer in the language they were asked

In our testing, when a prompt is in German, engines tend to lean on German-language sources: German sites, forums, review sites, and publisher roundups. English sources are used, but less, and a brand whose only presence is English tends to be named less often and described less confidently. The English site's quality does not transfer. Each market needs its own readable, quotable, consistent presence.

One source of truth, many renderings

Start with the canonical facts document: company description, product names, who each is for, pricing model, locations. Translate it once, carefully, with a human who knows the market, and treat each translation as the source of truth for that language.

Then every localized property copies from its translation: site, schema, llms.txt, profiles. The German Organization schema should say the same thing as the German homepage paragraph and the German OMR Reviews profile.

Where it goes wrong: the English site updates, the translations lag by six months, and the engine reads a contradiction between your German site and your English site. Put translation on the same change process as the English version.

Decide what stays the same

Some things should not localize: the company name, product names (unless you deliberately market under different names, which creates an entity problem you then need to manage), and the sameAs links in schema that tie every localized site to the same entity.

Some things must localize: pricing and currency, availability, support hours, legal entity and address, certifications that matter locally, integrations that are popular locally.

The things that stay the same let an engine understand that the German site and the English site are one company. The things that change let it answer the German buyer correctly.

Technical layer per market

• hreflang correctly set so crawlers know which page serves which language and region.
• Localized schema on localized pages, not English schema copied over. Description, price currency, and areaServed should be local.
• Crawler access checked per subdomain. Regional sites on separate infrastructure often have their own robots.txt and their own blocks.
• Rendering checked per market. Regional sites built later on a different framework sometimes render client-side when the main site does not.
• llms.txt at the root linking to each localized section, or one per localized domain, each generated from the translated facts.

Local sources are different sources

Run your prompt set in each language and record what gets cited. The sources will differ: different review aggregators, forums, Q&A sites, and publisher comparisons per market.

For each market, build a short source map: the three to five third-party properties the engine leans on. Then be present in each, legitimately: a claimed and accurate profile, honest reviews from real local customers, and a local team member answering questions in the local community.

This is where international programs tend to stall, because it is people work in each language and cannot be done from headquarters. Give each market owner a source map and the authority to work it.

Measurement per market

Translate the prompt set intent for intent, with a native speaker, because buyers phrase questions differently. Run it weekly per language and engine. Report share of answer per market.

The number that matters most is the gap between your strongest and weakest markets. A large gap usually means a lagging translation, a rendering problem on the regional site, or no local source presence.

Prioritize

Do not launch all markets at once. Pick the two where revenue is largest and share of answer is lowest. Fix those fully, then add markets one at a time using the same checklist.

First step

Run your 20 core prompts in the three languages that matter most and see where you are named and where you are not. At AEO GEO Labs that per-language baseline is the first thing we build for an international client, because it usually shows that the problem is concentrated in one or two markets rather than spread evenly. The free crawler and JavaScript checkers from AEO GEO Labs work on any subdomain, so test each regional site, not just the main one.