How AI legal research is changing cross-border work in Europe

Most AI legal tools were built for one job: reviewing a contract, redlining a clause, flagging risk in a document. Several do it well. But if you are advising on company law, civil obligations and regulatory duties in Germany, Poland, France and the Netherlands at the same time, that is a different task altogether. It is not redlining. It is knowing what the statute says in each jurisdiction, as it stands today, with a citation you can put your name to.

Cross-border work in Europe has always been research-heavy. The law is not hidden — finding it, reconciling it across systems and drafting from it is what eats the day.

Why contract tools fall short of multi-jurisdiction research

A contract tool is document-centric by design. You upload an agreement, the model finds deviations from a playbook, suggests edits and hands back a redline. For transactional work inside one legal system, that is genuinely useful.

Cross-border research runs the other way. You do not start from a document, you start from a question. Does a non-compete of this length survive under German, French and Dutch commercial law? What are the disclosure duties on an acquisition in all three? How does a European regulation sit against the local statute that implements it?

Answering that reliably takes the provision itself. Not a summary, not a paragraph lifted from a PDF that may have been amended since. The actual text, from the authoritative source, with a citation that says exactly where it came from.

Verification is the real bottleneck

Ask lawyers what slows a cross-border matter down and the answer is almost always the same: verification. Finding something that looks like an answer takes minutes. Confirming it is current, correctly cited and consistent across every jurisdiction involved takes the rest of the afternoon.

Generic retrieval makes this worse in a specific way. It pulls text chunks from a store and answers from whatever sits nearby, then points at a page offset rather than a legal unit. Amend the statute and the offset drifts. The answer still reads authoritative. For advice you are about to send, a citation you cannot check against the current text is not a citation — it is exposure.

How azta approaches it

azta is a research and drafting tool, not a redliner. It is built for the work where the answer has to be right and documented, across the company and commercial, contract and civil, and regulatory questions that European practice throws up.

Citations that resolve to the provision

In 17 of the 24 European jurisdictions it covers, azta runs on a tree-law engine: statutes are ingested as a structure — documents, units, attributes and the references between them — rather than as flat text. Researchers walk that tree, listing a provision’s children, reading a unit’s text and following a cross-reference to the unit it actually points at. Cross-references resolve deterministically instead of being guessed from surrounding text.

The consequence is that a citation resolves to a stable legal-unit identity rather than to a page offset, and renders as a link straight back to the provision. Coverage runs to 24 European jurisdictions in total, with new legislation, directives and official guidance indexed daily; the remaining seven are researched through the older flow, which is a real difference in depth and worth checking against the countries you actually advise on.

One question, several jurisdictions

A single request can carry several target countries. A dispatch-coverage gate then holds the answer back: every jurisdiction that surfaced a candidate law has to get a researcher actually sent into that law before the agent is allowed to reply. Finding a statute is not the same as reading it, and an answer assembled from search summaries alone is ungrounded — so the gate refuses it rather than letting a thin jurisdiction ride along on a well-researched one.

The working, not just the conclusion

Before an answer is published, a separate verifier audits it on four counts: whether the citation coverage is full, partial or insufficient; whether the answer contradicts itself; whether any cited provision has been amended recently enough to put the conclusion at temporal risk; and whether the answer is actually aligned with the scope of the question asked. It can send the agent back for more research, or ask you a clarifying question, instead of publishing.

The research tracker shows the working alongside that — what was searched, which sources were read, what was checked. Where jurisdictions diverge, you see the divergence rather than an averaged-out summary that hides it.

That is the part worth insisting on. In legal research a confident wrong answer is more dangerous than an acknowledged gap.

From research to the deliverable

Research is half the job. Once you have the statutory picture, someone still has to turn it into a memo the client reads or a partner signs. Canvas takes the research and drafts it as a side-panel artifact — a summary, an advisory memo or a client email — which the agent can then edit in place, with the citations carried through. Full editable Word documents are generated into the editor, and the finished deliverable exports to Word or PDF, or shares as a read-only link for review.

It also works inside your own material: uploaded PDFs (with real OCR), Word, Excel, PowerPoint, XML and Outlook .msg files are read and cited back with the exact passage highlighted in the original document, so a citation into a client file is as checkable as a citation into a statute.

For a firm whose bottleneck is producing the deliverable rather than finding the answer, that is where the hours come back.

What it means by practice area

Company and commercial

Formation requirements, director liability, shareholder rights and disclosure duties diverge across Europe even inside harmonised frameworks. A cross-border acquisition or a group restructuring turns on the local implementation, not the directive. That is the level the answer has to land at.

Contract and civil

Governing-law clauses, limitation periods, enforceability and remedies all sit in the local civil code. When a contract is governed by one system and performed in another you need both, side by side, with the points of conflict visible.

Regulatory

European frameworks are transposed unevenly. The directive is one thing; the national implementation, the regulator’s guidance and enforcement practice are another. For financial services, data protection, competition or environmental work, the research has to reach national level — which is exactly what searching 24 national databases by hand makes impractical.

Pricing and access

Enterprise legal AI is generally sold through a sales conversation, with pricing settled at firm level. azta Pro is €27 a month or €270 a year, self-serve, and the free tier gives five questions a month to test it on real matters first. Firms that need their own tenant get a dedicated single-tenant deployment — its own domain, its own database, SAML SSO and their own branding — but nobody has to go through that door to evaluate the product.

The Statute Finder is free to every signed-in user with no plan gate: a two-pane browser over the statute book with streaming search, a reading-first statute view, matched-unit highlighting and boolean filters. It is the cheapest way to see whether the law you rely on is actually in there, before any of the rest matters.

The shift is in the time, not the judgement

None of this replaces legal judgement. What the law means for this client in this situation still needs someone who understands the matter. What compresses is the distance between a question and a verified, citable answer — the retrieval, the reconciliation across systems, the first draft of the deliverable.

When a piece of research that used to take hours comes back cited in minutes, and the first draft comes with it, a small team can take on matters that used to belong to a much larger one. For cross-border practice in Europe, that is a structural change rather than a marginal one.