Sourced Answers, Not Confident Guesses

By now the cautionary tales are well known: lawyers sanctioned for briefs citing cases that did not exist, invented by a chatbot that sounded completely sure of itself. The lesson isn't "AI is useless for legal work." The lesson is narrower and more useful: in high-stakes filings, an answer you can't trace to a source is worse than no answer at all.

A confident guess costs you nothing to read and everything to rely on. The whole value of the work — in litigation and in patent prosecution alike — is that every assertion can be stood up against a real document.

Where the stakes actually live

Two kinds of practice make this especially sharp, and both are drowning in their own documents:

• 01

  Litigation. Building the case means knowing what's in tens of thousands of pages of discovery — and being able to point to the exact document and page behind every fact, while never crossing a privilege line.

• 02

  Patent prosecution. Answering an office action means reconciling the examiner's rejections with the claims, the specification, and prior art — much of which is already sitting in the firm's own files from related matters.

In both, the raw material for a strong answer usually already exists. It's just buried, and re-finding it by hand is where the hours and the risk pile up.

An answer that shows its work

This is the difference between a black box and a knowledge layer. A black box gives you a paragraph and asks you to trust it. A knowledge layer gives you the same paragraph and the receipt — the specific documents each part of the answer rests on, so a person can verify it before it ever reaches a filing.

That's the only responsible way to bring these tools into work where a wrong citation has consequences. The tool doesn't replace the attorney or the agent's judgment; it does the finding, and it shows exactly where every answer came from so the professional can do the deciding. We wrote more about that audit trail in The Chain of Custody.

The goal was never an AI that sounds right. It's a system that can prove it.