Legora builds legal research on an ontology and citator: what you must be able to check per workflow

Legora presents legal research built on a full ontology of law and an AI-native citator, with its own benchmark (BAR) that measures per answer whether claims carry a citation and whether those citations match the consulted sources. Judge such a tool on ontology scope, transparent grounding and governance per matter, not as a smarter search bar.

The occasion is a report by Artificial Lawyer of 7 September 2026 on tests by Harvey and Legora with OpenAI's GPT-6 Astra. In it, Artificial Lawyer describes how a Legora agent, powered by Astra, carries out a tie-out across 41 financial-summary documents, spots planted discrepancies and records every check with citations, with, according to the publication, roughly 40% better performance than the previous model GPT-5.6 Sol. Our editorial point: whether this works depends less on the model and more on how Legora treats citations — as objects within a structure, not as footnotes added afterwards.

What changes when Legora builds research on an ontology of law and an AI citator?

In the product description of Legora's agentic operating system, the company states that it rebuilds legal research with "comprehensive data, a full ontology of law, and an AI-native citator". Ontology and citator are therefore not extras, but the structural core of the research layer.

What that means in concrete terms, in our estimate:

• Sources become structured: case law, legislation, regulation and internal precedents are placed in an ontology rather than being searchable as loose text.
• The citator navigates that structure: it shows how authorities relate to each other (followed, distinguished, overruled) rather than only displaying a list of references.
• AI agents reason over the graph: an agent queries the structured representation of the law rather than matching flat text.

For those who rely on legal AI, the control question shifts as a result. No longer only "does the answer have a source?", but "is the scope of the underlying ontology appropriate and documented?".

How does Legora's BAR benchmark measure whether citations really support an answer?

In the Benchmark for Agentic Reasoning (BAR), Legora describes an evaluation that runs within a real matter document space. According to Legora, BAR measures two things separately: whether every claim in an AI answer carries a citation, and whether those citations match the sources the AI actually consulted — so-called grounding.

That distinction is important. A citation that is present is not the same as a citation that actually supports the answer. That citations can be misleading in practice is a known and measurable problem; we discussed this earlier in misleading AI citations and source checking in 2026. A benchmark that measures grounding separately from merely present references makes this difference explicitly testable.

Our analysis: here the integration of ontology and AI becomes verifiable. The citator does not only list sources, but is tested on the question of whether those sources support the answer. For users, the gain lies in the ability to use grounding scores as a threshold in decisions about deployment, in line with what we set out earlier for GPT-6 Astra in legal workflows and what you must demonstrate per task.

What does linking citations to Gemini Enterprise and Intapp mean for your governance?

In the announcement about Google Cloud's Gemini Enterprise for Legal, Legora reports a beta integration via the Model Context Protocol. According to Legora, a user can ask a legal question within Gemini, receive a text answer with links back to Legora threads, and then review the citations, sources, rights and legal context in Legora. This makes visible that governed legal research is not tied to a single supplier: just as Google links its Gemini environment here, other large providers such as Microsoft could play a comparable role with their own AI assistants as a front end above a structured research layer.

In addition, LawNext describes in the ILTACON report on the strategic partnership between Intapp and Legora that AI work in Legora respects firms' ethical walls, confidentiality and regulation, and that structured usage data go to Intapp Time.

Our interpretation of these two links together:

• Ontology-indexed citations become transferable between AI systems.
• That only works well if Legora remains the governance anchor for source review, rather than an external model — whether that is Gemini or a Microsoft assistant — flattening the citation structure into loose text.
• The link with ethical walls and time recording makes visible per matter which authorities an agent relied on and how that relates to access rules.

Which control questions should you ask per legal AI workflow?

For professionals working with confidential or high-risk information, we translate the above into three concrete questions per workflow:

1. Scope of the ontology: which jurisdictions, sources and internal knowledge are included, and is that delineation documented?
2. Insight into citations and grounding: can you inspect and export per answer which authorities support it, and how BAR-like grounding scores performed?
3. Preservation of structure in integrations: do links with systems such as Gemini and Intapp keep the ontology, citations and ethical-walls restrictions intact rather than flattening them?

Anyone who wants to answer these questions structurally arrives at a verification-focused approach. Our broader approach to that is set out in the topic hub on AI verification and checkable AI answers, and more specifically in our earlier piece on a verification layer above legal AI research tools.

In that verification perspective, a tool such as Vera can play a role as a console above such systems: it makes visible, per matter and workflow, which sources and verification steps underpinned an answer, and which corrections or disagreements between models occurred. Vera is not a chatbot and not its own language model; it does not promise correct answers and cannot remove the risk of hallucinations, but it does make control better possible by making verification steps visible. The professional final judgement remains with you. The editorial news value lies not in that verification product, but in the way Legora brings together ontology, AI and a native citator and makes that grounding measurable.

Sources: The article relies on Artificial Lawyer, Legora's own product and BAR pages, Legora's announcement about Gemini Enterprise for Legal and LawNext on the Intapp partnership.

← All articles in this topic ← All articles