Verification in Legal AI Is a Design Problem
Lawyers remain responsible. Legal AI should make its work easy to check before they sign.
The Complete Guide to Legal Fact Management
An introduction to legal fact management and how litigation teams turn scattered documents into a factual record that remains traceable to the evidence.
Legal fact management is the work of turning the documents in a matter into a factual record that lawyers can inspect, correct and reuse.
The term is useful because document storage and factual understanding solve different problems. A document management or eDiscovery system can hold and retrieve the files. The legal team still has to identify the people, events, dates, conflicts and gaps across them.
What legal fact management includes
The work includes extracting factual statements, preserving the source and surrounding context, normalizing people and dates, connecting related events across documents, recording contradictions and missing material, and allowing lawyers to correct the account.
A chronology presents one view of that record. The same reviewed facts can also support matter summaries, witness comparison, gap analysis, financial review and later drafting.
Why document search is not enough
Search is useful when the lawyer knows what to look for. Many litigation questions require the team to identify an event before it can formulate the right query.
A relevant fact may also be described with different words across an email, witness statement and attachment. Legal fact management creates a structured view across those sources while preserving the path back to the documents.
What a factual record should contain
A factual record should show what happened, who was involved, when it occurred, which document supports the statement, whether accounts conflict, whether expected material is missing and whether a lawyer has reviewed or corrected the fact.
The record should evolve as new documents arrive and corrections are made. The factual record is becoming infrastructure explains why that persistence becomes more valuable as firms use more AI tools.
Where AI helps
Language models can read unstructured documents and propose factual structure much faster than a manual first pass. That does not remove the need for review.
The system should show the exact source for each factual claim, report the scope of documents processed and preserve user corrections. The useful output is a record the legal team can inspect and carry into later work.
How to evaluate a system
Use a closed matter with a reference record reviewed by lawyers who know the file. Measure factual support, material omissions, source-review time, corrections required and whether later outputs use the corrected facts.
Avoid relying on a generic accuracy number without the task and scoring method behind it. Mary’s verification whitepaper provides a practical known-matter test.
Mary’s approach
Mary builds the factual record for litigation. It reads the matter documents, structures the facts and keeps each fact traceable to its source. Lawyers can then review the record and use it across chronologies and other work product.
The objective is to reduce repeated factual reconstruction while keeping the evidence visible. The National Compensation Lawyers case study shows how one firm used Mary to reduce time spent on document review.