Today, we’re introducing Ontological Legal Research, starting with the United States. It brings comprehensive legal data together with a proprietary ontology that maps how sources connect. This enables the Legora Agent to see the full picture of current law and navigate it with unprecedented accuracy.
One of the hardest problems in legal AI
AI fails at legal research in critical ways. Fabricated citations are the most visible failure. Even when the sources are real, AI can cite outdated law, misjudge a source’s authority, or find the main rule but miss the exception. It misses the full picture of current law.
Fixing this takes two things: comprehensive legal data and a way for the Agent to understand how the law connects.
Whatever it takes to get comprehensive data
The single most common frustration we hear from customers is that you can't get the best data and the best technology in one product. Legal publishers have comprehensive data but a weaker AI layer. AI providers have advanced technology but not yet comprehensive data for legal research. Too often, the data our clients buy access to is simply not available in the AI platforms they want to use — one prominent publisher CEO recently noted that only 1% of their data was actually available in a partner's platform.
So we’re bringing comprehensive legal data into Legora. Primary law is a public asset, but getting hold of it is surprisingly difficult. The challenges vary by jurisdiction and source type. In Germany, only about 3% of case law is digital. In the UK, legislation from official sources is often two to four weeks behind. In the US, the court reporters are printed by publishers who won’t sell a digital copy. We partner wherever it enables us to deliver the best possible user experience, but where a partnership won’t deliver the right quality or coverage, we go directly to the source and get it ourselves: manual requests, physical scanning. Whatever it takes.
US case law is one example. The courts are clear that judicial opinions are a public asset (we brought in outside counsel to confirm), but publishers won’t sell a digital feed of citable cases. So we collect slip opinions directly from court websites within 24-48 hours, then get the physical advance sheets and bound volumes within about a week of release. We take the spines off, remove publisher-added material such as headnotes and annotations, then scan them. To get to 99.997% accuracy, three separate people check every word on every page.
Our Attorney Editors bring decades of experience at traditional publishers. They map source coverage in each jurisdiction by document type, including required historical coverage and update frequency. Customer requests drive priorities, and our customers' law librarians help audit the comprehensiveness and accuracy of our data.
Building a map of current law
With the data in place, the second challenge is understanding how the sources connect, and what is current law. US legal sources alone span tens of millions of documents, far more than any AI model can process at once. The model depends on search to select the material it will read.
“For legal research, semantic search is like throwing darts blindfolded"
- Arvid Winterfeldt, Product Manager, Legal Research
The industry standard uses keyword and semantic search to find relevant passages for the model to reason over. “For legal research, semantic search is like throwing darts blindfolded. It picks up opinions that look useful in isolation, but misses critical context. A later decision may have overruled, narrowed, or criticized one of those opinions. If that decision is never retrieved, the model doesn’t have the information it needs to understand how the law has changed,” says Arvid Winterfeldt, Product Manager for Legal Research at Legora and former CEO at Qura, acquired by Legora earlier this year.
A better model alone won’t solve that problem. It needs a map of connections between sources.
So we build a proprietary ontology: a map of how the law connects. We map defined terms in the statutes to the cases that interpret them. If a case has been overruled, we connect it to the decision that overruled it, so whenever the Legora Agent uses the earlier case, it sees the later decision too. We connect rules to their exceptions. We map the authority of legal sources, from persuasive commentary to binding Supreme Court precedent. We build these relationships in advance, so the Agent can use them on every research run.
We evaluate everything we build against thousands of legal research questions. Alongside dramatic improvements in performance and accuracy, the ontological data structure makes the Agent 40% faster. It follows a structured map of the law from the start instead of reconstructing those relationships on every run.
This technology comes from Qura, which Legora acquired in April. The team spent three years building it for civil law jurisdictions, where some of the leading firms in the Nordics adopted Qura as their sole legal research platform. We’re now applying that approach to common law.
Next: the citator
The ontology maps the relationships between legal sources. Our AI-native citator verifies whether the legal propositions in those sources are still good law, and defines those relationships for a specific legal proposition, instead of the whole document.
Legal publishers have spent 150 years doing this work by hand. Our Attorney Editors set the standards, make the difficult calls and verify the output. AI lets us apply that expertise at a scale that was previously impossible.
More information about the citator is coming later this year.
Ontological Legal Research is live in the US, with more jurisdictions coming soon.
