A legal AI product can become multilingual without acquiring equivalent legal capability in each jurisdiction. International expansion therefore raises a separate question: “What legal information does a product need in order to support another legal system?”

The internationalisation of legal AI is often discussed in terms of language. “Can the model understand German?”, “Can it draft in French?”, “Can it search Spanish documents and answer in English?”

These questions matter, but language coverage and jurisdictional coverage are different things. A model may operate very effectively in several languages while still depending on the legal information made available to it for a particular jurisdiction. Supporting German as a language does not by itself determine which German legal sources are covered, which concepts the product recognises, how local terminology is normalised, how historical changes are handled or how German concepts relate to concepts familiar to an English-speaking user.

This distinction becomes important as legal-AI products expand internationally. Adding a jurisdiction requires a series of choices about legal information. Which areas of law matter for the intended workflows? What level of source coverage is sufficient? Which legal concepts should be represented explicitly? How should terminology, definitions and relationships be organised? What requires local expert validation? How will changes be maintained?

The answers to such questions will differ by product. A litigation research tool needs different coverage from a contract-drafting assistant. A product used primarily by local lawyers may need less comparative mediation than one designed for cross-border work.

This is why coverage should be treated as a product and data-strategy question rather than a flag added to a language selector. The product team needs to decide what users should reasonably be able to do in the jurisdiction and then work backwards to the legal information required to support those workflows.

For cross-border applications, the relationships between jurisdictions become important as well. A user may ask for the German equivalent of a familiar English concept, compare contractual mechanisms or research how several systems address the same commercial problem. The system needs enough information to recognise where concepts align and where the comparison becomes more complicated.

Some of this can be inferred dynamically from source material. Some relationships are sufficiently important and reusable that making them explicit can improve consistency. The strategic question is where that boundary should sit.

Internationalisation therefore creates a legal-data architecture problem. Documents, retrieval, models and local expertise all remain necessary, but they need to be organised around a clear view of what “supporting” a jurisdiction actually means.

As legal AI moves into more countries, this distinction is likely to become increasingly visible. Multilingual capability helps the system communicate across languages. Jurisdictional capability determines how well it can work with the law behind the languages.